1. Position your content as a thought leader

Competitive intelligence from SERP analysis allows you to know what stakeholders expect to find. Without knowing the competition, you need luck to differentiate or distinguish your content.

2. Keyword expansion and SEO strategy

SERPs are very useful for SEO. SERP analytics help you to find low competition keywords. Machine learning natural language processing on search engine rankings and improve the quality of your keyword research. And all of that in a standardized, highly automated workflow, with low human effort.

3. Find content ideas

Mine headlines, keyphrases and keywords from SERPs to generate content ideas and extract insights about the interests of your audience beyond what search engine data suggests directly.

docker run -d -p 4445:4444 -p 5901:5900 selenium/standalone-chrome-debug:latest

## 45c8aa7cbb24c56707e230d9ad2cd35ad94ba35203daab8fee9f2910f5d4d0d3

pacman::p_load(RSelenium, rvest, Rcrawler, tidyverse, data.table, tidytable, tidytext, reticulate, twinning, fastrtext, wordVectors, udpipe)



#get the selenium server running


remDr <- remoteDriver(
  remoteServerAddr = "localhost",
  port = 4445L,
  browserName = "chrome"
)

Sys.sleep(1)
remDr$open(silent = TRUE)

startpage_url <- function(search_terms){
  
  search_terms <- search_terms
  query <- str_replace_all(search_terms, " ", "%20")
  url <- paste0("https://www.startpage.com/sp/search?query=", query)

  url
  
}

q <- "how to generate document embeddings"
google_this <- startpage_url(q)


remDr$navigate(google_this)


Sys.sleep(30)

# extract the raw html
get_html <- function(remDr){
  remDr$getPageSource() %>% .[[1]] %>% read_html()
}


#css selectors

paginate_button <- "button.next"
result_link <- "a.w-gl__result-url"
result_title <- "div.w-gl__result-second-line-container"
result_description <- "p.w-gl__description"



#loop through and collect search results

search_results_all <- data.table(title_result=c(), description_result=c(), link_result=c())


check_next <- function(x){remDr$findElement("css", paginate_button)}
next_button_check <- possibly(check_next, otherwise = NA)
next_button_present <- next_button_check()
  


while (!is.na(next_button_present)){
  
  
  
  title <- remDr %>% get_html() %>% html_nodes(result_title) %>% html_text()
  description <-  remDr %>% get_html() %>% html_nodes(result_description) %>% html_text()
  link <-  remDr %>% get_html() %>% html_nodes(result_link) %>% html_text()
  
  
  search_results <- data.table(title_result=title, description_result=description, link_result=link)
  search_results_all <- bind_rows(search_results_all, search_results)
  
  
  next_button <- remDr$findElement("css", paginate_button)
  
  
  next_button$clickElement()
  next_button_present <- next_button_check()
  Sys.sleep(10)
  
  
}



search_results_all %>% kableExtra::kable()

search_results_all_distinct <- search_results_all %>% distinct.(link_result) %>% filter(str_detect(link_result, "stackoverflow", negate = T))
search_results_all_distinct$link <- str_squish(search_results_all_distinct$link_result)

#scrape the search results

search_result_scrape_all <- data.table(search_engine_rank=c(), source=c(), element=c(), value=c())


#No image loading
prefs = list("profile.managed_default_content_settings.images" = 2L)
cprof <- list(chromeOptions = list(prefs = prefs))
  
  
remDr <- remoteDriver(
  remoteServerAddr = "localhost",
  port = 4445L,
  browserName = "chrome", extraCapabilities = cprof
  )

scrape_search_results <- function(link){
  
    remDr$open(silent = T)
    remDr$navigate(link)
    raw_html <- remDr %>% get_html() %>% as.character()
    remDr$close()
  
  
  
  
  search_result_scrape <- Rcrawler::ContentScraper(HTmlText = raw_html, CssPatterns = c("p", "h1", "h2", "body"), PatternsName = c("paragraphs", "title", "subtitle", "body"), ManyPerPattern=T, asDataFrame=T) %>% rownames_to_column(var = "element") %>% melt(id.vars="element") %>% distinct.(-variable)
  


  search_result_scrape_all <- search_result_scrape_all %>% bind_rows.(search_result_scrape)
  
  
  
  Sys.sleep(1)

  
  search_result_scrape_all
  
}

scr_safe <- possibly(scrape_search_results, otherwise = data.table(search_engine_rank=c(), source=c(), element=c(), value=c()))

search_results_data <- map(search_results_all_distinct$link_result, .f = scr_safe)

search_results_data2 <- search_results_data %>% bind_rows.(.id = T) %>% rename(rank=.id)

search_results_data2 %>% sample_n(10) %>% kableExtra::kable()

containers <- system("docker ps", intern = TRUE)
container_filter <- system("docker ps", intern = TRUE) %>% str_detect("selenium/standalone-chrome-debug")

container_id <- containers[container_filter] %>% str_split(" ")
container_id <- container_id[[1]][1]


system(paste0("docker stop ", container_id), intern = T)

## [1] "45c8aa7cbb24"

Finding relevant long tail keywords in scraped SERPs

An obvious strong signal for long tail keyword phrases are ngrams in the headlines used in SERPs. First extract all ngrams:

headlines <- search_results_data2 %>% filter(element%in%c("title", "subtitle")) %>% na.omit()

ngrams <- tidytext::unnest_ngrams(headlines, output = "ngram", input = "value", n = 6, n_min = 3) %>% distinct.(ngram) %>% na.omit()

ngrams <- ngrams$ngram

print("Wanna glimpse at a compression of this? Have some patience, human intellect. I'm not ready yet.")

## [1] "Wanna glimpse at a compression of this? Have some patience, human intellect. I'm not ready yet."

Next let’s transform them:

from sentence_transformers import SentenceTransformer, util
model = SentenceTransformer('all-distilroberta-v1')

#Our sentences we like to encode
sentences = r.ngrams
query = r.q

#Sentences are encoded by calling model.encode()
embeddings_ngrams = model.encode(sentences)
embeddings_query= model.encode(r.q)


#Compute cosine-similarits
cosine_scores = util.cos_sim(embeddings_query, embeddings_ngrams)

Compress the dataset into 50 sementically representative records, but random sampling is just not rotating our shapes intensively enough…

Let’s extract 50 representative records which are statistically similar to all other records using the Twinning algorithm:

ngram_transformed <- py$embeddings_ngrams %>% as.data.frame()
rownames(ngram_transformed) <- ngrams

extract_summarizing_records <- function(n, embeddings){
  r <- (nrow(embeddings)/n) 
  r <- round(r+2*sign(r)*.Machine$double.eps)
  sample <- twin(embeddings, r = r)
  sample
  
}


ngrams[extract_summarizing_records(50, ngram_transformed)] %>% kableExtra::kable()

x
mover’s embedding from word2vec to
word mover’s embedding from word2vec
from word2vec to document embedding
embedding from word2vec
word and document embeddings for
how to develop word embeddings in
word embeddings in
word mover’s embedding from
from word embeddings to document
word embeddings to document distances
word embeddings in python with
embeddings in python with gensim
in python with
computation and language
how to develop
experimental projects with community
arxivlabs experimental projects with community collaborators
and search tools
and data associated
code and data associated
a guide to building document
guide to building document embeddings part
building document embeddings part 2
6 document embeddings for
document embeddings for rankings
embeddings for rankings and
recommenders and search
tables and topics
figures tables and topics from
topic detection and tracking
detection and tracking with time
and tracking with time
time aware document
the law using document
of the law
spatial representations of the law using
vectors spatial representations of the
case vectors spatial representations
document embeddings part
embeddings for arabic sentiment
embeddings for sentiment analysis
for arabic sentiment analysis
from this paper
tutorial 5 document
no full text available
title:generating word and document embeddings
looking for the
bibliographic and citation tools
the evaluation metric
preparing your download

Distracted again by those pretty shapes.

Let’s get the top n most similar long tail keyword suggestions for our original query:

similarities <- data.table(query=q, ngram=ngrams, similarity=py$cosine_scores$tolist()[[1]]) %>% arrange(desc(similarity))

similarities %>% head(100) %>% kableExtra::kable()

query	ngram	similarity
how to generate document embeddings	to building document embeddings	0.8747796
how to generate document embeddings	building document embeddings	0.8548772
how to generate document embeddings	using document embeddings	0.8442889
how to generate document embeddings	to document embedding	0.8320625
how to generate document embeddings	to building document embeddings part	0.8307059
how to generate document embeddings	to building document embeddings part 2	0.8286431
how to generate document embeddings	guide to building document embeddings	0.8237830
how to generate document embeddings	document embeddings for	0.8235297
how to generate document embeddings	embeddings to document	0.8125913
how to generate document embeddings	building document embeddings part	0.8105815
how to generate document embeddings	guide to building document embeddings part	0.8080642
how to generate document embeddings	a guide to building document embeddings	0.7991555
how to generate document embeddings	building document embeddings part 2	0.7933071
how to generate document embeddings	document embeddings part	0.7752839
how to generate document embeddings	document embeddings part 2	0.7620034
how to generate document embeddings	tutorial 5 document embeddings	0.7604587
how to generate document embeddings	5 document embeddings	0.7600654
how to generate document embeddings	aware document embeddings	0.7537373
how to generate document embeddings	word and document embeddings for	0.7521918
how to generate document embeddings	from word embeddings to document	0.7489915
how to generate document embeddings	6 document embeddings for	0.7457391
how to generate document embeddings	6 document embeddings	0.7443236
how to generate document embeddings	chapter 6 document embeddings	0.7272525
how to generate document embeddings	word and document embeddings	0.7244108
how to generate document embeddings	chapter 6 document embeddings for	0.7199622
how to generate document embeddings	and document embeddings	0.7116172
how to generate document embeddings	word embeddings to document	0.7094321
how to generate document embeddings	and document embeddings for	0.6923491
how to generate document embeddings	document embeddings for rankings and	0.6787176
how to generate document embeddings	title:generating word and document embeddings	0.6620581
how to generate document embeddings	title:generating word and document embeddings for	0.6548300
how to generate document embeddings	from word embeddings	0.6429227
how to generate document embeddings	with time aware document embeddings	0.6414276
how to generate document embeddings	6 document embeddings for rankings and	0.6339916
how to generate document embeddings	develop word embeddings	0.6333530
how to generate document embeddings	time aware document embeddings	0.6311547
how to generate document embeddings	document embeddings for sentiment analysis	0.6293886
how to generate document embeddings	how to develop word embeddings in	0.6268539
how to generate document embeddings	document embeddings for arabic	0.6267931
how to generate document embeddings	from word2vec to document embedding	0.6266655
how to generate document embeddings	how to develop word embeddings	0.6255153
how to generate document embeddings	document embeddings for rankings	0.6255038
how to generate document embeddings	and document embeddings for sentiment analysis	0.6204889
how to generate document embeddings	embeddings to document distances	0.6157715
how to generate document embeddings	develop word embeddings in	0.6147120
how to generate document embeddings	from word embeddings to document distances	0.6120901
how to generate document embeddings	embedding from word2vec to document embedding	0.6100764
how to generate document embeddings	from word embeddings to	0.6013952
how to generate document embeddings	chapter 6 document embeddings for rankings	0.5985473
how to generate document embeddings	document embeddings for rankings and recommendations	0.5916984
how to generate document embeddings	6 document embeddings for rankings	0.5859146
how to generate document embeddings	document embeddings for sentiment	0.5804358
how to generate document embeddings	word2vec to document embedding	0.5779676
how to generate document embeddings	embedding from word2vec to document	0.5769333
how to generate document embeddings	law using document embeddings	0.5727847
how to generate document embeddings	and document embeddings for sentiment	0.5684689
how to generate document embeddings	word embeddings tutorial	0.5680511
how to generate document embeddings	word embeddings to document distances	0.5436367
how to generate document embeddings	document embeddings for arabic sentiment analysis	0.5427437
how to generate document embeddings	to develop word embeddings	0.5413585
how to generate document embeddings	to develop word embeddings in	0.5389535
how to generate document embeddings	develop word embeddings in python with	0.5374320
how to generate document embeddings	develop word embeddings in python	0.5330083
how to generate document embeddings	document embeddings for arabic sentiment	0.5290453
how to generate document embeddings	embedding from word2vec	0.5288824
how to generate document embeddings	embeddings for arabic	0.5273871
how to generate document embeddings	word and document embeddings for sentiment	0.5254203
how to generate document embeddings	to develop word embeddings in python	0.5249420
how to generate document embeddings	the law using document embeddings	0.5221492
how to generate document embeddings	word embeddings in	0.5184467
how to generate document embeddings	word embeddings to	0.5133757
how to generate document embeddings	embeddings in python with	0.5126769
how to generate document embeddings	of the law using document embeddings	0.5068490
how to generate document embeddings	embeddings in python	0.5066578
how to generate document embeddings	embeddings part 2	0.5029533
how to generate document embeddings	to building document	0.4991868
how to generate document embeddings	embedding from word2vec to	0.4965441
how to generate document embeddings	mover’s embedding from word2vec to document	0.4871210
how to generate document embeddings	mover’s embedding from word2vec to document	0.4855789
how to generate document embeddings	embeddings for rankings and	0.4839105
how to generate document embeddings	word embeddings in python with	0.4775580
how to generate document embeddings	word embeddings in python	0.4725776
how to generate document embeddings	guide to building document	0.4636505
how to generate document embeddings	bert word embeddings tutorial	0.4636116
how to generate document embeddings	a guide to building document	0.4616931
how to generate document embeddings	embeddings for rankings	0.4496233
how to generate document embeddings	embeddings for sentiment analysis	0.4495095
how to generate document embeddings	word mover’s embedding from word2vec	0.4367592
how to generate document embeddings	tracking with time aware document embeddings	0.4257493
how to generate document embeddings	word mover’s embedding from word2vec to	0.4206680
how to generate document embeddings	word mover’s embedding from word2vec	0.4199873
how to generate document embeddings	word mover’s embedding from	0.4199831
how to generate document embeddings	embeddings for arabic sentiment analysis	0.4165690
how to generate document embeddings	title:generating word and document	0.4163545
how to generate document embeddings	word mover’s embedding	0.4142923
how to generate document embeddings	embeddings for rankings and recommendations	0.4121665
how to generate document embeddings	embeddings for arabic sentiment	0.4113144
how to generate document embeddings	embeddings for sentiment	0.4080955
how to generate document embeddings	mover’s embedding from word2vec	0.4076054
how to generate document embeddings	word embeddings in python with gensim	0.4062557

There might also be idiosyncratic expressions not represented by the learnings of our transformer model. Let’s train an ngram-fasttext model for that purpose on this corpus:

texts <- search_results_data2 %>% filter(element%in%c("title", "subtitle", "paragraphs")) %>% na.omit()

texts <- tolower(texts$value)
tmp_file_txt <- tempfile()
tmp_file_model <- "/home/knut/Documents/doc_embs_fasttext.model"
writeLines(text = texts, con = tmp_file_txt)
execute(commands = c("skipgram", "-input", tmp_file_txt, "-output", tmp_file_model, "-verbose", 2, "-wordNgrams", 5, "-maxn", 10, "-epoch", 2000, "-thread", 15))

## 
Read 0M words
## Number of words:  627
## Number of labels: 0
## 
Progress:   0.4% words/sec/thread:  117105 lr:  0.049806 avg.loss:  2.661299 ETA:   0h 0m25s
Progress:   0.8% words/sec/thread:  117974 lr:  0.049609 avg.loss:  2.468449 ETA:   0h 0m25s
Progress:   1.2% words/sec/thread:  118110 lr:  0.049413 avg.loss:  2.344289 ETA:   0h 0m25s
Progress:   1.6% words/sec/thread:  117983 lr:  0.049219 avg.loss:  2.236573 ETA:   0h 0m25s
Progress:   2.0% words/sec/thread:  118387 lr:  0.049020 avg.loss:  2.146931 ETA:   0h 0m25s
Progress:   2.4% words/sec/thread:  118395 lr:  0.048825 avg.loss:  2.116807 ETA:   0h 0m24s
Progress:   2.7% words/sec/thread:  118205 lr:  0.048631 avg.loss:  2.094915 ETA:   0h 0m24s
Progress:   3.1% words/sec/thread:  118360 lr:  0.048434 avg.loss:  2.071258 ETA:   0h 0m24s
Progress:   3.5% words/sec/thread:  118607 lr:  0.048234 avg.loss:  2.050964 ETA:   0h 0m24s
Progress:   3.9% words/sec/thread:  118460 lr:  0.048041 avg.loss:  2.042554 ETA:   0h 0m24s
Progress:   4.3% words/sec/thread:  118463 lr:  0.047845 avg.loss:  2.020189 ETA:   0h 0m24s
Progress:   4.7% words/sec/thread:  118319 lr:  0.047652 avg.loss:  2.007978 ETA:   0h 0m24s
Progress:   5.1% words/sec/thread:  118456 lr:  0.047453 avg.loss:  1.995940 ETA:   0h 0m24s
Progress:   5.5% words/sec/thread:  118447 lr:  0.047258 avg.loss:  1.981349 ETA:   0h 0m24s
Progress:   5.9% words/sec/thread:  118347 lr:  0.047064 avg.loss:  1.970432 ETA:   0h 0m24s
Progress:   6.3% words/sec/thread:  118388 lr:  0.046868 avg.loss:  1.966935 ETA:   0h 0m23s
Progress:   6.6% words/sec/thread:  118171 lr:  0.046678 avg.loss:  1.959670 ETA:   0h 0m23s
Progress:   7.0% words/sec/thread:  118133 lr:  0.046484 avg.loss:  1.950206 ETA:   0h 0m23s
Progress:   7.4% words/sec/thread:  118192 lr:  0.046287 avg.loss:  1.944411 ETA:   0h 0m23s
Progress:   7.8% words/sec/thread:  117976 lr:  0.046098 avg.loss:  1.940148 ETA:   0h 0m23s
Progress:   8.2% words/sec/thread:  117893 lr:  0.045906 avg.loss:  1.931570 ETA:   0h 0m23s
Progress:   8.6% words/sec/thread:  117822 lr:  0.045714 avg.loss:  1.925169 ETA:   0h 0m23s
Progress:   8.9% words/sec/thread:  117653 lr:  0.045526 avg.loss:  1.923746 ETA:   0h 0m23s
Progress:   9.3% words/sec/thread:  117716 lr:  0.045329 avg.loss:  1.922762 ETA:   0h 0m23s
Progress:   9.7% words/sec/thread:  117632 lr:  0.045137 avg.loss:  1.917217 ETA:   0h 0m23s
Progress:  10.1% words/sec/thread:  117579 lr:  0.044945 avg.loss:  1.914724 ETA:   0h 0m23s
Progress:  10.5% words/sec/thread:  117608 lr:  0.044750 avg.loss:  1.910344 ETA:   0h 0m23s
Progress:  10.9% words/sec/thread:  117524 lr:  0.044559 avg.loss:  1.903496 ETA:   0h 0m22s
Progress:  11.3% words/sec/thread:  117541 lr:  0.044364 avg.loss:  1.897073 ETA:   0h 0m22s
Progress:  11.7% words/sec/thread:  117484 lr:  0.044173 avg.loss:  1.892617 ETA:   0h 0m22s
Progress:  12.0% words/sec/thread:  117428 lr:  0.043981 avg.loss:  1.886137 ETA:   0h 0m22s
Progress:  12.4% words/sec/thread:  117381 lr:  0.043790 avg.loss:  1.883704 ETA:   0h 0m22s
Progress:  12.8% words/sec/thread:  117332 lr:  0.043598 avg.loss:  1.878408 ETA:   0h 0m22s
Progress:  13.2% words/sec/thread:  117303 lr:  0.043406 avg.loss:  1.874179 ETA:   0h 0m22s
Progress:  13.6% words/sec/thread:  117325 lr:  0.043211 avg.loss:  1.875709 ETA:   0h 0m22s
Progress:  14.0% words/sec/thread:  117245 lr:  0.043021 avg.loss:  1.872937 ETA:   0h 0m22s
Progress:  14.3% words/sec/thread:  117222 lr:  0.042829 avg.loss:  1.866312 ETA:   0h 0m22s
Progress:  14.7% words/sec/thread:  117279 lr:  0.042632 avg.loss:  1.865677 ETA:   0h 0m22s
Progress:  15.1% words/sec/thread:  117230 lr:  0.042441 avg.loss:  1.867309 ETA:   0h 0m21s
Progress:  15.5% words/sec/thread:  117208 lr:  0.042249 avg.loss:  1.866775 ETA:   0h 0m21s
Progress:  15.9% words/sec/thread:  117200 lr:  0.042055 avg.loss:  1.867217 ETA:   0h 0m21s
Progress:  16.3% words/sec/thread:  117092 lr:  0.041869 avg.loss:  1.868941 ETA:   0h 0m21s
Progress:  16.6% words/sec/thread:  117084 lr:  0.041676 avg.loss:  1.870448 ETA:   0h 0m21s
Progress:  17.0% words/sec/thread:  117053 lr:  0.041485 avg.loss:  1.871866 ETA:   0h 0m21s
Progress:  17.4% words/sec/thread:  117048 lr:  0.041292 avg.loss:  1.872645 ETA:   0h 0m21s
Progress:  17.8% words/sec/thread:  117061 lr:  0.041097 avg.loss:  1.874170 ETA:   0h 0m21s
Progress:  18.2% words/sec/thread:  117042 lr:  0.040905 avg.loss:  1.874849 ETA:   0h 0m21s
Progress:  18.6% words/sec/thread:  117059 lr:  0.040710 avg.loss:  1.877104 ETA:   0h 0m21s
Progress:  19.0% words/sec/thread:  117041 lr:  0.040518 avg.loss:  1.877936 ETA:   0h 0m20s
Progress:  19.4% words/sec/thread:  117042 lr:  0.040324 avg.loss:  1.878894 ETA:   0h 0m20s
Progress:  19.7% words/sec/thread:  117041 lr:  0.040131 avg.loss:  1.882261 ETA:   0h 0m20s
Progress:  20.1% words/sec/thread:  116993 lr:  0.039941 avg.loss:  1.882429 ETA:   0h 0m20s
Progress:  20.5% words/sec/thread:  116969 lr:  0.039750 avg.loss:  1.881168 ETA:   0h 0m20s
Progress:  20.9% words/sec/thread:  116975 lr:  0.039556 avg.loss:  1.883036 ETA:   0h 0m20s
Progress:  21.3% words/sec/thread:  116925 lr:  0.039367 avg.loss:  1.886035 ETA:   0h 0m20s
Progress:  21.6% words/sec/thread:  116886 lr:  0.039178 avg.loss:  1.887735 ETA:   0h 0m20s
Progress:  22.0% words/sec/thread:  116916 lr:  0.038982 avg.loss:  1.887925 ETA:   0h 0m20s
Progress:  22.4% words/sec/thread:  116861 lr:  0.038794 avg.loss:  1.887774 ETA:   0h 0m20s
Progress:  22.8% words/sec/thread:  116854 lr:  0.038601 avg.loss:  1.886975 ETA:   0h 0m19s
Progress:  23.2% words/sec/thread:  116891 lr:  0.038404 avg.loss:  1.886806 ETA:   0h 0m19s
Progress:  23.6% words/sec/thread:  116846 lr:  0.038215 avg.loss:  1.886063 ETA:   0h 0m19s
Progress:  24.0% words/sec/thread:  116841 lr:  0.038023 avg.loss:  1.885102 ETA:   0h 0m19s
Progress:  24.3% words/sec/thread:  116869 lr:  0.037827 avg.loss:  1.884412 ETA:   0h 0m19s
Progress:  24.7% words/sec/thread:  116848 lr:  0.037636 avg.loss:  1.882732 ETA:   0h 0m19s
Progress:  25.1% words/sec/thread:  116832 lr:  0.037444 avg.loss:  1.880893 ETA:   0h 0m19s
Progress:  25.5% words/sec/thread:  116839 lr:  0.037250 avg.loss:  1.879765 ETA:   0h 0m19s
Progress:  25.9% words/sec/thread:  116840 lr:  0.037057 avg.loss:  1.879006 ETA:   0h 0m19s
Progress:  26.3% words/sec/thread:  116817 lr:  0.036866 avg.loss:  1.876567 ETA:   0h 0m19s
Progress:  26.7% words/sec/thread:  116833 lr:  0.036671 avg.loss:  1.874629 ETA:   0h 0m19s
Progress:  27.0% words/sec/thread:  116850 lr:  0.036476 avg.loss:  1.875475 ETA:   0h 0m18s
Progress:  27.4% words/sec/thread:  116770 lr:  0.036292 avg.loss:  1.875180 ETA:   0h 0m18s
Progress:  27.8% words/sec/thread:  116727 lr:  0.036104 avg.loss:  1.874745 ETA:   0h 0m18s
Progress:  28.2% words/sec/thread:  116753 lr:  0.035908 avg.loss:  1.873551 ETA:   0h 0m18s
Progress:  28.6% words/sec/thread:  116702 lr:  0.035722 avg.loss:  1.871960 ETA:   0h 0m18s
Progress:  28.9% words/sec/thread:  116714 lr:  0.035527 avg.loss:  1.869789 ETA:   0h 0m18s
Progress:  29.3% words/sec/thread:  116764 lr:  0.035328 avg.loss:  1.868881 ETA:   0h 0m18s
Progress:  29.7% words/sec/thread:  116715 lr:  0.035141 avg.loss:  1.867670 ETA:   0h 0m18s
Progress:  30.1% words/sec/thread:  116726 lr:  0.034947 avg.loss:  1.865761 ETA:   0h 0m18s
Progress:  30.5% words/sec/thread:  116731 lr:  0.034753 avg.loss:  1.865076 ETA:   0h 0m18s
Progress:  30.9% words/sec/thread:  116718 lr:  0.034562 avg.loss:  1.865134 ETA:   0h 0m17s
Progress:  31.3% words/sec/thread:  116701 lr:  0.034367 avg.loss:  1.863878 ETA:   0h 0m17s
Progress:  31.7% words/sec/thread:  116715 lr:  0.034172 avg.loss:  1.863231 ETA:   0h 0m17s
Progress:  32.0% words/sec/thread:  116727 lr:  0.033977 avg.loss:  1.863965 ETA:   0h 0m17s
Progress:  32.4% words/sec/thread:  116663 lr:  0.033793 avg.loss:  1.863570 ETA:   0h 0m17s
Progress:  32.8% words/sec/thread:  116696 lr:  0.033596 avg.loss:  1.863031 ETA:   0h 0m17s
Progress:  33.2% words/sec/thread:  116689 lr:  0.033404 avg.loss:  1.863507 ETA:   0h 0m17s
Progress:  33.6% words/sec/thread:  116657 lr:  0.033215 avg.loss:  1.863521 ETA:   0h 0m17s
Progress:  34.0% words/sec/thread:  116671 lr:  0.033021 avg.loss:  1.863606 ETA:   0h 0m17s
Progress:  34.3% words/sec/thread:  116672 lr:  0.032827 avg.loss:  1.863864 ETA:   0h 0m17s
Progress:  34.7% words/sec/thread:  116650 lr:  0.032638 avg.loss:  1.863950 ETA:   0h 0m16s
Progress:  35.1% words/sec/thread:  116657 lr:  0.032444 avg.loss:  1.863386 ETA:   0h 0m16s
Progress:  35.5% words/sec/thread:  116643 lr:  0.032253 avg.loss:  1.863859 ETA:   0h 0m16s
Progress:  35.9% words/sec/thread:  116626 lr:  0.032063 avg.loss:  1.863409 ETA:   0h 0m16s
Progress:  36.3% words/sec/thread:  116597 lr:  0.031874 avg.loss:  1.862776 ETA:   0h 0m16s
Progress:  36.6% words/sec/thread:  116591 lr:  0.031683 avg.loss:  1.862992 ETA:   0h 0m16s
Progress:  37.0% words/sec/thread:  116604 lr:  0.031488 avg.loss:  1.863580 ETA:   0h 0m16s
Progress:  37.4% words/sec/thread:  116564 lr:  0.031301 avg.loss:  1.862702 ETA:   0h 0m16s
Progress:  37.8% words/sec/thread:  116568 lr:  0.031108 avg.loss:  1.862596 ETA:   0h 0m16s
Progress:  38.2% words/sec/thread:  116577 lr:  0.030914 avg.loss:  1.862939 ETA:   0h 0m16s
Progress:  38.5% words/sec/thread:  116553 lr:  0.030725 avg.loss:  1.862549 ETA:   0h 0m15s
Progress:  38.9% words/sec/thread:  116575 lr:  0.030529 avg.loss:  1.862132 ETA:   0h 0m15s
Progress:  39.3% words/sec/thread:  116565 lr:  0.030338 avg.loss:  1.862645 ETA:   0h 0m15s
Progress:  39.7% words/sec/thread:  116544 lr:  0.030149 avg.loss:  1.862558 ETA:   0h 0m15s
Progress:  40.1% words/sec/thread:  116561 lr:  0.029953 avg.loss:  1.862140 ETA:   0h 0m15s
Progress:  40.5% words/sec/thread:  116540 lr:  0.029764 avg.loss:  1.862109 ETA:   0h 0m15s
Progress:  40.9% words/sec/thread:  116533 lr:  0.029573 avg.loss:  1.861749 ETA:   0h 0m15s
Progress:  41.2% words/sec/thread:  116532 lr:  0.029380 avg.loss:  1.861537 ETA:   0h 0m15s
Progress:  41.6% words/sec/thread:  116524 lr:  0.029189 avg.loss:  1.862069 ETA:   0h 0m15s
Progress:  42.0% words/sec/thread:  116536 lr:  0.028994 avg.loss:  1.861417 ETA:   0h 0m15s
Progress:  42.4% words/sec/thread:  116525 lr:  0.028803 avg.loss:  1.860970 ETA:   0h 0m14s
Progress:  42.8% words/sec/thread:  116529 lr:  0.028610 avg.loss:  1.861345 ETA:   0h 0m14s
Progress:  43.2% words/sec/thread:  116524 lr:  0.028418 avg.loss:  1.861034 ETA:   0h 0m14s
Progress:  43.5% words/sec/thread:  116503 lr:  0.028229 avg.loss:  1.860393 ETA:   0h 0m14s
Progress:  43.9% words/sec/thread:  116503 lr:  0.028037 avg.loss:  1.860635 ETA:   0h 0m14s
Progress:  44.3% words/sec/thread:  116494 lr:  0.027846 avg.loss:  1.860481 ETA:   0h 0m14s
Progress:  44.7% words/sec/thread:  116487 lr:  0.027654 avg.loss:  1.859146 ETA:   0h 0m14s
Progress:  45.1% words/sec/thread:  116479 lr:  0.027463 avg.loss:  1.858490 ETA:   0h 0m14s
Progress:  45.5% words/sec/thread:  116498 lr:  0.027267 avg.loss:  1.857463 ETA:   0h 0m14s
Progress:  45.8% words/sec/thread:  116469 lr:  0.027080 avg.loss:  1.853353 ETA:   0h 0m14s
Progress:  46.2% words/sec/thread:  116456 lr:  0.026890 avg.loss:  1.851278 ETA:   0h 0m13s
Progress:  46.6% words/sec/thread:  116440 lr:  0.026700 avg.loss:  1.850621 ETA:   0h 0m13s
Progress:  47.0% words/sec/thread:  116387 lr:  0.026515 avg.loss:  1.849012 ETA:   0h 0m13s
Progress:  47.4% words/sec/thread:  116407 lr:  0.026318 avg.loss:  1.847662 ETA:   0h 0m13s
Progress:  47.7% words/sec/thread:  116372 lr:  0.026133 avg.loss:  1.846003 ETA:   0h 0m13s
Progress:  48.1% words/sec/thread:  116334 lr:  0.025948 avg.loss:  1.844317 ETA:   0h 0m13s
Progress:  48.5% words/sec/thread:  116359 lr:  0.025751 avg.loss:  1.843601 ETA:   0h 0m13s
Progress:  48.9% words/sec/thread:  116284 lr:  0.025573 avg.loss:  1.842059 ETA:   0h 0m13s
Progress:  49.2% words/sec/thread:  116282 lr:  0.025381 avg.loss:  1.842254 ETA:   0h 0m13s
Progress:  49.6% words/sec/thread:  116292 lr:  0.025187 avg.loss:  1.839707 ETA:   0h 0m13s
Progress:  50.0% words/sec/thread:  116238 lr:  0.025006 avg.loss:  1.837136 ETA:   0h 0m13s
Progress:  50.4% words/sec/thread:  116229 lr:  0.024816 avg.loss:  1.835751 ETA:   0h 0m12s
Progress:  50.7% words/sec/thread:  116214 lr:  0.024627 avg.loss:  1.834370 ETA:   0h 0m12s
Progress:  51.1% words/sec/thread:  116183 lr:  0.024442 avg.loss:  1.834000 ETA:   0h 0m12s
Progress:  51.5% words/sec/thread:  116182 lr:  0.024250 avg.loss:  1.833244 ETA:   0h 0m12s
Progress:  51.9% words/sec/thread:  116168 lr:  0.024056 avg.loss:  1.832729 ETA:   0h 0m12s
Progress:  52.3% words/sec/thread:  116134 lr:  0.023872 avg.loss:  1.831303 ETA:   0h 0m12s
Progress:  52.6% words/sec/thread:  116097 lr:  0.023688 avg.loss:  1.830850 ETA:   0h 0m12s
Progress:  53.0% words/sec/thread:  116082 lr:  0.023499 avg.loss:  1.829580 ETA:   0h 0m12s
Progress:  53.4% words/sec/thread:  116043 lr:  0.023316 avg.loss:  1.827873 ETA:   0h 0m12s
Progress:  53.7% words/sec/thread:  116003 lr:  0.023134 avg.loss:  1.826242 ETA:   0h 0m12s
Progress:  54.1% words/sec/thread:  116012 lr:  0.022940 avg.loss:  1.825626 ETA:   0h 0m11s
Progress:  54.5% words/sec/thread:  115970 lr:  0.022757 avg.loss:  1.824465 ETA:   0h 0m11s
Progress:  54.9% words/sec/thread:  115946 lr:  0.022571 avg.loss:  1.824333 ETA:   0h 0m11s
Progress:  55.2% words/sec/thread:  115954 lr:  0.022378 avg.loss:  1.823980 ETA:   0h 0m11s
Progress:  55.6% words/sec/thread:  115909 lr:  0.022197 avg.loss:  1.822485 ETA:   0h 0m11s
Progress:  56.0% words/sec/thread:  115885 lr:  0.022011 avg.loss:  1.820361 ETA:   0h 0m11s
Progress:  56.4% words/sec/thread:  115855 lr:  0.021809 avg.loss:  1.817954 ETA:   0h 0m11s
Progress:  56.8% words/sec/thread:  115826 lr:  0.021624 avg.loss:  1.817111 ETA:   0h 0m11s
Progress:  57.1% words/sec/thread:  115804 lr:  0.021438 avg.loss:  1.816064 ETA:   0h 0m11s
Progress:  57.5% words/sec/thread:  115764 lr:  0.021257 avg.loss:  1.815614 ETA:   0h 0m11s
Progress:  57.9% words/sec/thread:  115768 lr:  0.021065 avg.loss:  1.816395 ETA:   0h 0m11s
Progress:  58.2% words/sec/thread:  115726 lr:  0.020881 avg.loss:  1.815572 ETA:   0h 0m10s
Progress:  58.6% words/sec/thread:  115695 lr:  0.020698 avg.loss:  1.814022 ETA:   0h 0m10s
Progress:  59.0% words/sec/thread:  115702 lr:  0.020505 avg.loss:  1.813037 ETA:   0h 0m10s
Progress:  59.4% words/sec/thread:  115660 lr:  0.020324 avg.loss:  1.811471 ETA:   0h 0m10s
Progress:  59.7% words/sec/thread:  115636 lr:  0.020139 avg.loss:  1.810544 ETA:   0h 0m10s
Progress:  60.1% words/sec/thread:  115641 lr:  0.019940 avg.loss:  1.808708 ETA:   0h 0m10s
Progress:  60.5% words/sec/thread:  115583 lr:  0.019763 avg.loss:  1.807727 ETA:   0h 0m10s
Progress:  60.9% words/sec/thread:  115579 lr:  0.019573 avg.loss:  1.806656 ETA:   0h 0m10s
Progress:  61.2% words/sec/thread:  115589 lr:  0.019380 avg.loss:  1.805446 ETA:   0h 0m10s
Progress:  61.6% words/sec/thread:  115544 lr:  0.019200 avg.loss:  1.804588 ETA:   0h 0m10s
Progress:  62.0% words/sec/thread:  115528 lr:  0.019013 avg.loss:  1.803563 ETA:   0h 0m 9s
Progress:  62.4% words/sec/thread:  115523 lr:  0.018824 avg.loss:  1.802682 ETA:   0h 0m 9s
Progress:  62.7% words/sec/thread:  115498 lr:  0.018639 avg.loss:  1.801162 ETA:   0h 0m 9s
Progress:  63.1% words/sec/thread:  115467 lr:  0.018457 avg.loss:  1.799667 ETA:   0h 0m 9s
Progress:  63.5% words/sec/thread:  115475 lr:  0.018264 avg.loss:  1.798565 ETA:   0h 0m 9s
Progress:  63.8% words/sec/thread:  115440 lr:  0.018081 avg.loss:  1.797894 ETA:   0h 0m 9s
Progress:  64.2% words/sec/thread:  115404 lr:  0.017900 avg.loss:  1.796863 ETA:   0h 0m 9s
Progress:  64.6% words/sec/thread:  115424 lr:  0.017703 avg.loss:  1.796571 ETA:   0h 0m 9s
Progress:  65.0% words/sec/thread:  115401 lr:  0.017519 avg.loss:  1.795551 ETA:   0h 0m 9s
Progress:  65.3% words/sec/thread:  115354 lr:  0.017341 avg.loss:  1.794479 ETA:   0h 0m 9s
Progress:  65.7% words/sec/thread:  115360 lr:  0.017149 avg.loss:  1.793751 ETA:   0h 0m 8s
Progress:  66.1% words/sec/thread:  115331 lr:  0.016966 avg.loss:  1.792482 ETA:   0h 0m 8s
Progress:  66.4% words/sec/thread:  115314 lr:  0.016780 avg.loss:  1.791495 ETA:   0h 0m 8s
Progress:  66.8% words/sec/thread:  115333 lr:  0.016584 avg.loss:  1.790495 ETA:   0h 0m 8s
Progress:  67.2% words/sec/thread:  115296 lr:  0.016404 avg.loss:  1.789320 ETA:   0h 0m 8s
Progress:  67.6% words/sec/thread:  115273 lr:  0.016220 avg.loss:  1.788199 ETA:   0h 0m 8s
Progress:  67.9% words/sec/thread:  115276 lr:  0.016029 avg.loss:  1.787403 ETA:   0h 0m 8s
Progress:  68.3% words/sec/thread:  115235 lr:  0.015850 avg.loss:  1.786334 ETA:   0h 0m 8s
Progress:  68.7% words/sec/thread:  115226 lr:  0.015662 avg.loss:  1.785471 ETA:   0h 0m 8s
Progress:  69.1% words/sec/thread:  115225 lr:  0.015472 avg.loss:  1.784188 ETA:   0h 0m 8s
Progress:  69.4% words/sec/thread:  115215 lr:  0.015284 avg.loss:  1.783526 ETA:   0h 0m 8s
Progress:  69.8% words/sec/thread:  115192 lr:  0.015101 avg.loss:  1.782694 ETA:   0h 0m 7s
Progress:  70.2% words/sec/thread:  115184 lr:  0.014913 avg.loss:  1.782173 ETA:   0h 0m 7s
Progress:  70.5% words/sec/thread:  115172 lr:  0.014726 avg.loss:  1.781780 ETA:   0h 0m 7s
Progress:  70.9% words/sec/thread:  115146 lr:  0.014543 avg.loss:  1.780979 ETA:   0h 0m 7s
Progress:  71.3% words/sec/thread:  115150 lr:  0.014352 avg.loss:  1.780273 ETA:   0h 0m 7s
Progress:  71.7% words/sec/thread:  115126 lr:  0.014169 avg.loss:  1.779668 ETA:   0h 0m 7s
Progress:  72.0% words/sec/thread:  115090 lr:  0.013990 avg.loss:  1.778561 ETA:   0h 0m 7s
Progress:  72.4% words/sec/thread:  115113 lr:  0.013792 avg.loss:  1.777818 ETA:   0h 0m 7s
Progress:  72.8% words/sec/thread:  115098 lr:  0.013606 avg.loss:  1.776707 ETA:   0h 0m 7s
Progress:  73.1% words/sec/thread:  115055 lr:  0.013429 avg.loss:  1.775059 ETA:   0h 0m 7s
Progress:  73.5% words/sec/thread:  115070 lr:  0.013234 avg.loss:  1.774380 ETA:   0h 0m 6s
Progress:  73.9% words/sec/thread:  115041 lr:  0.013053 avg.loss:  1.773524 ETA:   0h 0m 6s
Progress:  74.3% words/sec/thread:  115035 lr:  0.012865 avg.loss:  1.772976 ETA:   0h 0m 6s
Progress:  74.7% words/sec/thread:  115043 lr:  0.012672 avg.loss:  1.772104 ETA:   0h 0m 6s
Progress:  75.0% words/sec/thread:  115011 lr:  0.012492 avg.loss:  1.771179 ETA:   0h 0m 6s
Progress:  75.4% words/sec/thread:  115004 lr:  0.012304 avg.loss:  1.770301 ETA:   0h 0m 6s
Progress:  75.8% words/sec/thread:  114995 lr:  0.012108 avg.loss:  1.769217 ETA:   0h 0m 6s
Progress:  76.2% words/sec/thread:  114981 lr:  0.011922 avg.loss:  1.768617 ETA:   0h 0m 6s
Progress:  76.5% words/sec/thread:  114970 lr:  0.011736 avg.loss:  1.768169 ETA:   0h 0m 6s
Progress:  76.9% words/sec/thread:  114951 lr:  0.011552 avg.loss:  1.767480 ETA:   0h 0m 6s
Progress:  77.3% words/sec/thread:  114954 lr:  0.011361 avg.loss:  1.767134 ETA:   0h 0m 5s
Progress:  77.6% words/sec/thread:  114941 lr:  0.011175 avg.loss:  1.765572 ETA:   0h 0m 5s
Progress:  78.0% words/sec/thread:  114922 lr:  0.010992 avg.loss:  1.764724 ETA:   0h 0m 5s
Progress:  78.4% words/sec/thread:  114922 lr:  0.010801 avg.loss:  1.764313 ETA:   0h 0m 5s
Progress:  78.8% words/sec/thread:  114894 lr:  0.010621 avg.loss:  1.762537 ETA:   0h 0m 5s
Progress:  79.1% words/sec/thread:  114876 lr:  0.010437 avg.loss:  1.761785 ETA:   0h 0m 5s
Progress:  79.5% words/sec/thread:  114878 lr:  0.010247 avg.loss:  1.760372 ETA:   0h 0m 5s
Progress:  79.9% words/sec/thread:  114852 lr:  0.010066 avg.loss:  1.759619 ETA:   0h 0m 5s
Progress:  80.2% words/sec/thread:  114849 lr:  0.009877 avg.loss:  1.759029 ETA:   0h 0m 5s
Progress:  80.6% words/sec/thread:  114832 lr:  0.009693 avg.loss:  1.758268 ETA:   0h 0m 5s
Progress:  81.0% words/sec/thread:  114821 lr:  0.009506 avg.loss:  1.757874 ETA:   0h 0m 5s
Progress:  81.4% words/sec/thread:  114808 lr:  0.009321 avg.loss:  1.757017 ETA:   0h 0m 4s
Progress:  81.7% words/sec/thread:  114798 lr:  0.009135 avg.loss:  1.756669 ETA:   0h 0m 4s
Progress:  82.1% words/sec/thread:  114790 lr:  0.008948 avg.loss:  1.756217 ETA:   0h 0m 4s
Progress:  82.5% words/sec/thread:  114776 lr:  0.008763 avg.loss:  1.755239 ETA:   0h 0m 4s
Progress:  82.9% words/sec/thread:  114784 lr:  0.008570 avg.loss:  1.754797 ETA:   0h 0m 4s
Progress:  83.2% words/sec/thread:  114767 lr:  0.008387 avg.loss:  1.753966 ETA:   0h 0m 4s
Progress:  83.6% words/sec/thread:  114750 lr:  0.008203 avg.loss:  1.753143 ETA:   0h 0m 4s
Progress:  84.0% words/sec/thread:  114752 lr:  0.008013 avg.loss:  1.752713 ETA:   0h 0m 4s
Progress:  84.3% words/sec/thread:  114740 lr:  0.007827 avg.loss:  1.751804 ETA:   0h 0m 4s
Progress:  84.7% words/sec/thread:  114731 lr:  0.007641 avg.loss:  1.751293 ETA:   0h 0m 4s
Progress:  85.1% words/sec/thread:  114732 lr:  0.007451 avg.loss:  1.750643 ETA:   0h 0m 3s
Progress:  85.5% words/sec/thread:  114714 lr:  0.007267 avg.loss:  1.749682 ETA:   0h 0m 3s
Progress:  85.8% words/sec/thread:  114719 lr:  0.007076 avg.loss:  1.749181 ETA:   0h 0m 3s
Progress:  86.2% words/sec/thread:  114712 lr:  0.006889 avg.loss:  1.748418 ETA:   0h 0m 3s
Progress:  86.6% words/sec/thread:  114692 lr:  0.006706 avg.loss:  1.747698 ETA:   0h 0m 3s
Progress:  87.0% words/sec/thread:  114682 lr:  0.006521 avg.loss:  1.747315 ETA:   0h 0m 3s
Progress:  87.3% words/sec/thread:  114670 lr:  0.006336 avg.loss:  1.746411 ETA:   0h 0m 3s
Progress:  87.7% words/sec/thread:  114667 lr:  0.006147 avg.loss:  1.746137 ETA:   0h 0m 3s
Progress:  88.1% words/sec/thread:  114650 lr:  0.005964 avg.loss:  1.745614 ETA:   0h 0m 3s
Progress:  88.4% words/sec/thread:  114644 lr:  0.005777 avg.loss:  1.744758 ETA:   0h 0m 3s
Progress:  88.8% words/sec/thread:  114640 lr:  0.005589 avg.loss:  1.744041 ETA:   0h 0m 2s
Progress:  89.2% words/sec/thread:  114624 lr:  0.005405 avg.loss:  1.743205 ETA:   0h 0m 2s
Progress:  89.6% words/sec/thread:  114618 lr:  0.005217 avg.loss:  1.742629 ETA:   0h 0m 2s
Progress:  89.9% words/sec/thread:  114611 lr:  0.005030 avg.loss:  1.742063 ETA:   0h 0m 2s
Progress:  90.3% words/sec/thread:  114589 lr:  0.004849 avg.loss:  1.740690 ETA:   0h 0m 2s
Progress:  90.7% words/sec/thread:  114597 lr:  0.004657 avg.loss:  1.740221 ETA:   0h 0m 2s
Progress:  91.1% words/sec/thread:  114587 lr:  0.004462 avg.loss:  1.739545 ETA:   0h 0m 2s
Progress:  91.4% words/sec/thread:  114566 lr:  0.004281 avg.loss:  1.738865 ETA:   0h 0m 2s
Progress:  91.8% words/sec/thread:  114577 lr:  0.004087 avg.loss:  1.738246 ETA:   0h 0m 2s
Progress:  92.2% words/sec/thread:  114571 lr:  0.003900 avg.loss:  1.737370 ETA:   0h 0m 2s
Progress:  92.6% words/sec/thread:  114554 lr:  0.003717 avg.loss:  1.736695 ETA:   0h 0m 1s
Progress:  93.0% words/sec/thread:  114562 lr:  0.003525 avg.loss:  1.736048 ETA:   0h 0m 1s
Progress:  93.3% words/sec/thread:  114546 lr:  0.003342 avg.loss:  1.735231 ETA:   0h 0m 1s
Progress:  93.7% words/sec/thread:  114532 lr:  0.003158 avg.loss:  1.734913 ETA:   0h 0m 1s
Progress:  94.1% words/sec/thread:  114536 lr:  0.002967 avg.loss:  1.734135 ETA:   0h 0m 1s
Progress:  94.4% words/sec/thread:  114525 lr:  0.002782 avg.loss:  1.733564 ETA:   0h 0m 1s
Progress:  94.8% words/sec/thread:  114505 lr:  0.002601 avg.loss:  1.733147 ETA:   0h 0m 1s
Progress:  95.2% words/sec/thread:  114495 lr:  0.002415 avg.loss:  1.732504 ETA:   0h 0m 1s
Progress:  95.5% words/sec/thread:  114453 lr:  0.002244 avg.loss:  1.731961 ETA:   0h 0m 1s
Progress:  95.9% words/sec/thread:  114448 lr:  0.002057 avg.loss:  1.730890 ETA:   0h 0m 1s
Progress:  96.3% words/sec/thread:  114444 lr:  0.001869 avg.loss:  1.730098 ETA:   0h 0m 0s
Progress:  96.6% words/sec/thread:  114440 lr:  0.001681 avg.loss:  1.729531 ETA:   0h 0m 0s
Progress:  97.0% words/sec/thread:  114440 lr:  0.001492 avg.loss:  1.728660 ETA:   0h 0m 0s
Progress:  97.4% words/sec/thread:  114443 lr:  0.001301 avg.loss:  1.727964 ETA:   0h 0m 0s
Progress:  97.8% words/sec/thread:  114439 lr:  0.001114 avg.loss:  1.727339 ETA:   0h 0m 0s
Progress:  98.2% words/sec/thread:  114441 lr:  0.000924 avg.loss:  1.726579 ETA:   0h 0m 0s
Progress:  98.5% words/sec/thread:  114443 lr:  0.000734 avg.loss:  1.725957 ETA:   0h 0m 0s
Progress:  98.9% words/sec/thread:  114444 lr:  0.000544 avg.loss:  1.725470 ETA:   0h 0m 0s
Progress:  99.3% words/sec/thread:  114437 lr:  0.000357 avg.loss:  1.724744 ETA:   0h 0m 0s
Progress:  99.7% words/sec/thread:  114439 lr:  0.000167 avg.loss:  1.724161 ETA:   0h 0m 0s
Progress: 100.0% words/sec/thread:  114391 lr: -0.000001 avg.loss:  1.723721 ETA:   0h 0m 0s
Progress: 100.0% words/sec/thread:  114390 lr:  0.000000 avg.loss:  1.723721 ETA:   0h 0m 0s

model <- fastrtext::load_model(tmp_file_model)


ngramsfasttext <- get_sentence_representation(model, ngrams) %>% as.VectorSpaceModel()
rownames(ngramsfasttext) <- ngrams


query_ft <- get_sentence_representation(model, q) %>% as.VectorSpaceModel()
rownames(query_ft) <- q


similar_ngrams_ft <-nearest_to(ngramsfasttext, query_ft, 200) 


similar_ngrams_ft%>% kableExtra::kable()

	x
embeddings to document	0.1228207
embeddings to document distances	0.1309724
title:generating word and document embeddings	0.1377801
title:generating word and document	0.1405947
word embeddings to document	0.1530124
word embeddings to document distances	0.1547942
from word embeddings to document distances	0.1655688
title:generating word and document embeddings for	0.1659545
from word embeddings to document	0.1730659
to building document embeddings	0.1734497
to document distances	0.1867444
and document embeddings	0.1875613
to document embedding	0.1908590
how to develop word embeddings	0.2023478
word and document embeddings	0.2031054
word and document	0.2075009
to building document embeddings part	0.2075886
a guide to building document embeddings	0.2123483
how to develop word embeddings in	0.2151130
guide to building document embeddings	0.2199200
and document embeddings for	0.2222097
word and document embeddings for	0.2259280
document embeddings for rankings and	0.2306636
guide to building document embeddings part	0.2308323
title:generating word and	0.2308880
word embeddings to	0.2335062
document embeddings for	0.2362187
from word2vec to document	0.2364173
mover’s embedding from word2vec to document	0.2415065
from word embeddings to	0.2416003
from word2vec to document embedding	0.2440056
embedding from word2vec to document	0.2440056
to building document embeddings part 2	0.2444263
word2vec to document	0.2462507
word2vec to document embedding	0.2482768
the law using document embeddings	0.2505018
document embeddings for rankings and recommendations	0.2507367
word and document embeddings for sentiment	0.2523996
mover’s embedding from word2vec to document	0.2546550
and document embeddings for sentiment	0.2584987
to building document	0.2622567
to develop word embeddings	0.2624329
document embeddings for rankings	0.2634049
of the law using document embeddings	0.2644605
6 document embeddings for rankings and	0.2671340
how to develop word	0.2677881
to develop word embeddings in	0.2693108
using document embeddings	0.2794391
embedding from word2vec to document embedding	0.2830800
document embeddings for sentiment	0.2920219
to develop word embeddings in python	0.2923414
document embeddings for arabic	0.2949671
chapter 6 document embeddings for	0.2953914
and document embeddings for sentiment analysis	0.3008579
tracking with time aware document embeddings	0.3022142
a guide to building document	0.3048720
6 document embeddings for	0.3058429
6 document embeddings for rankings	0.3079802
chapter 6 document embeddings for rankings	0.3108288
aware document embeddings	0.3141646
building document embeddings	0.3142161
with time aware document embeddings	0.3146568
embeddings for rankings and	0.3168887
word mover’s embedding from word2vec to	0.3172669
law using document embeddings	0.3179776
document embeddings part	0.3184815
how to develop	0.3190012
the law using document	0.3230650
mover’s embedding from word2vec to	0.3279785
to develop word	0.3281045
of the law using document	0.3284245
word mover’s embedding from word2vec to	0.3284722
document embeddings for arabic sentiment	0.3298279
embeddings for rankings and recommendations	0.3361477
word embeddings in	0.3382552
time aware document embeddings	0.3385937
word embeddings in python with	0.3397420
embedding from word2vec to	0.3409970
and tracking with time aware document	0.3425569
representations of the law using document	0.3445011
mover’s embedding from word2vec to	0.3454412
guide to building document	0.3460415
document embeddings for sentiment analysis	0.3465597
building document embeddings part	0.3471218
word embeddings in python	0.3495636
chapter 6 document embeddings	0.3496228
word embeddings in python with gensim	0.3508270
document embeddings part 2	0.3510619
from word2vec to	0.3544031
develop word embeddings in	0.3590375
6 document embeddings	0.3598216
from word embeddings	0.3608725
embeddings in python with	0.3621338
develop word embeddings in python with	0.3647269
looking for the full text	0.3697036
embeddings for rankings	0.3715321
document embeddings for arabic sentiment analysis	0.3724344
building document embeddings part 2	0.3781274
embeddings in python	0.3783573
embeddings in python with gensim	0.3785495
develop word embeddings in python	0.3839654
looking for the	0.3842144
for the full text	0.3892750
bert word embeddings	0.3922282
5 document embeddings	0.3959679
and tracking with	0.3960179
looking for the full	0.3971563
the full text	0.4016959
for rankings and	0.4049541
develop word embeddings	0.4049937
word embeddings tutorial	0.4051417
word mover’s embedding from	0.4054255
bert word embeddings tutorial	0.4073938
case vectors spatial representations of the	0.4087541
a guide to	0.4105952
tracking with time aware document	0.4106590
a guide to building	0.4126731
tutorial 5 document embeddings	0.4131265
and tracking with time	0.4144062
word mover’s embedding from	0.4174591
word mover’s embedding from word2vec	0.4220108
embeddings for arabic	0.4249423
vectors spatial representations of the	0.4261178
recommenders and search	0.4273539
figures tables and topics from this	0.4276156
tables and topics from this	0.4298384
for rankings and recommendations	0.4316614
and data associated with this	0.4332093
and citation tools	0.4332287
tables and topics from this paper	0.4336068
for the full	0.4355414
word mover’s embedding from word2vec	0.4356163
code and data associated with this	0.4360946
law using document	0.4371090
computer science computation and	0.4385289
recommenders and search tools	0.4404017
with time aware document	0.4407333
and tracking with time aware	0.4407894
embeddings for sentiment	0.4413947
and topics from this	0.4420748
word mover’s embedding	0.4444898
representations of the	0.4463178
spatial representations of the	0.4501823
science computation and	0.4505381
bibliographic and citation tools	0.4523700
detection and tracking with	0.4525358
and topics from this paper	0.4525856
with community collaborators	0.4541014
word mover’s embedding	0.4559564
and data associated with this article	0.4563103
detection and tracking with time	0.4586392
vectors spatial representations of the law	0.4601051
mover’s embedding from	0.4612503
of the law using	0.4619586
and data associated with	0.4623081
mover’s embedding from word2vec	0.4628096
bibliographic and citation	0.4649743
code and data associated with	0.4652127
representations of the law using	0.4661506
of the law	0.4666836
embeddings for arabic sentiment	0.4668968
computer science computation and language	0.4684936
figures tables and topics from	0.4688777
figures tables and	0.4706526
representations of the law	0.4708312
tables and topics from	0.4712852
spatial representations of the law using	0.4719104
and search tools	0.4743817
detection and tracking	0.4798411
in python with	0.4813077
spatial representations of the law	0.4813859
embeddings for sentiment analysis	0.4818796
mover’s embedding from	0.4825808
rankings and recommendations	0.4842365
mover’s embedding from word2vec	0.4846669
designing the evaluation	0.4871351
and topics from	0.4872253
and data associated	0.4881795
detection and tracking with time aware	0.4883160
embeddings part 2	0.4914431
code and data associated	0.4918750
case vectors spatial representations of	0.4923790
in python with gensim	0.4934976
the law using	0.4956175
embeddings for arabic sentiment analysis	0.4963960
science computation and language	0.4970849
experimental projects with community collaborators	0.4974520
code and data	0.4981775
computation and language	0.4982093
guide to building	0.5015119
embedding from word2vec	0.5017209
from this paper	0.5049279
projects with community collaborators	0.5052224
arxivlabs experimental projects with community collaborators	0.5059232
associated with this	0.5064865
time aware document	0.5095086
data associated with this	0.5102675
topic detection and tracking with time	0.5108899
tracking with time	0.5140204
topic detection and tracking with	0.5186386

These look alright, but we still can extract better candidates to save us from asking the keyword volume api too often and also extend the queries to include synonyms. Let’s first get the verb phrases and questions from these query candidates.

flairmodels <- import("flair.models")


multitagger <- flairmodels$MultiTagger
tagger <- multitagger$load(c("pos"))

flairdata <- import("flair.data")

#why is it so hard to comprehend for those Parseltonguers that methods and for loops are wordcel code whilst functions and tables are pure zen 

pos_tagger_tidy <- function(sentence){
  sents <- flairdata$Sentence(sentence)
  tagger$predict(sentences = sents)
  tags <- sents$to_tagged_string() %>% data.table(raw=.) %>% separate.(raw, c("sen", "tags"), sep = "\\[") %>% mutate(tags=str_remove(tags, "\\]"))
  tags <- cbind(tags, setDT(tags)[, tstrsplit(tags, ", ")]) %>% select.(-tags) %>% melt(id.vars="sen") %>% separate.(value, c("token", "upos"), sep = "/") %>% mutate(token=str_remove_all(token, '"')) %>% select.(-sen, -variable)
  tags
  
}


all_phrases <- c(similarities$ngram, rownames(similar_ngrams_ft)) %>% unique()


all_pos_tags <- purrr::map(all_phrases, purrrogress::with_progress(pos_tagger_tidy)) %>% data.table::rbindlist(idcol = T)

Let’s take a look at the verbphrases:

extract_verbphrases <- function(sentence){
  
  pos_tags <- pos_tagger_tidy(sentence)
  pos_tags$phrase_tag <- as_phrasemachine(pos_tags$upos, type = "penn-treebank")
  verbphrase <- keywords_phrases(pos_tags$phrase_tag, term = pos_tags$token, 
                                pattern = "((A|N)*N(P+D*(A|N)*N)*P*(M|V)*V(M|V)*|(M|V)*V(M|V)*D*(A|N)*N(P+D*(A|N)*N)*|(M|V)*V(M|V)*(P+D*(A|N)*N)+|(A|N)*N(P+D*(A|N)*N)*P*((M|V)*V(M|V)*D*(A|N)*N(P+D*(A|N)*N)*|(M|V)*V(M|V)*(P+D*(A|N)*N)+))", is_regex = TRUE, 
                                ngram_max = 4, 
                                detailed = T) %>% slice_tail.(1)

  verbphrase

  
}



all_verbphrases <- purrr::map(all_phrases, purrrogress::with_progress(extract_verbphrases)) %>% data.table::rbindlist(idcol = T)%>% distinct.(keyword) %>% na.omit()




all_verbphrases  %>% kableExtra::kable()

keyword
using document embeddings
document embedding
building document embeddings
generating word
develop word embeddings
embedding from word2vec
tracking with time
mover ’s embedding
word2vec to document
develop word
using document
document distances
data associated
associated with this article
law using
is semantic scholar
designing the evaluation
designing the evaluation metric
add a result
benchmarks add

Let’s also take a look at phrases containing questions:

question_ids <- all_pos_tags %>% filter(upos%in%c("WDT", "WP", "WP$", "WRB")) %>% select.(.id)


questions <- all_phrases[question_ids$.id] 

questions %>% kableExtra::kable()

x
how to develop word embeddings in
how to develop word embeddings
how to develop word
how to develop
what is semantic scholar
what is semantic

To find out relevant long tail query volumes, we will also combine verbs, questions and synonyms for document embeddings.

#find some synonyms for the main term and nouns

nouns <- all_pos_tags %>% filter.(upos%in%c("NN", "NNP", "NNPS", "NNS"))
unigram_nouns <- nouns %>% distinct.(token)

noun_lemmas <- udpipe(unigram_nouns$token, object = udpipe_load_model("/home/knut/Documents/english-ewt-ud-2.5-191206.udpipe")) %>% distinct(lemma)



ngram_nouns <- nouns %>% 
     group_by(.id) %>% 
     summarise(text = str_c(token, collapse = " ")) %>% distinct.(text)


nouns <- bind_rows.(unigram_nouns, ngram_nouns %>% rename(token=text)) %>% distinct.()

nouns %>% kableExtra::kable()

token
building
document
embeddings
embedding
part
guide
tutorial
word
chapter
rankings
title
time
sentiment
analysis
arabic
word2vec
distances
recommendations
law
python
mover
gensim
aware
representations
code
data
tables
topics
figures
citation
tools
article
vectors
paper
topic
detection
text
community
collaborators
recommenders
search
case
tracking
projects
scholar
science
computation
language
evaluation
arxivlabs
metric
computer
result
download
tokens
benchmarks
flairnlp
flair
building document embeddings
document embeddings
building document embeddings part
guide building document embeddings
embeddings document
guide building document embeddings part
guide document embeddings
document embeddings part
tutorial document embeddings
word document embeddings
word embeddings document
chapter document embeddings
document embeddings rankings
title word document embeddings
word embeddings
time document embeddings
document embeddings sentiment analysis
document embeddings arabic
word2vec document embedding
embeddings document distances
word embeddings document distances
chapter document embeddings rankings
document embeddings rankings recommendations
document embeddings sentiment
word2vec embedding
word2vec document
law document embeddings
word embeddings tutorial
word embeddings python
embeddings arabic
word document embeddings sentiment
embeddings python
embeddings part
building document
mover word2vec document
mover embedding word2vec document
embeddings rankings
guide building document
embeddings sentiment analysis
word mover word2vec
word mover embedding word2vec
word mover
title word document
embeddings rankings recommendations
embeddings sentiment
mover word2vec
word embeddings python gensim
mover embedding word2vec
embeddings python gensim
word mover embedding
mover embedding
time document
word document
time aware document
representations law document
title word
law document
code data
tables topics
figures tables topics
citation tools
data article
vectors representations
topic detection
community collaborators
topics paper
recommenders search tools
search tools
representations law
recommenders search
sentiment analysis
case vectors representations
guide building
topic detection tracking
topic detection tracking time
python gensim
chapter document
rankings recommendations
tables topics paper
projects community
vectors representations law
science computation language
figures tables
arxivlabs projects community
projects community collaborators
arxivlabs projects
evaluation metric
computer science computation language
arxivlabs projects community collaborators
case vectors
computer science computation
science computation
computation language
detection tracking time
detection tracking
benchmarks result
flairnlp flair

Now extract the verbs.

verbs <- all_pos_tags %>% filter.(upos%in%c("VB", "VBD", "VBG", "VBN", "VBP", "VBZ"))%>% distinct.(token)


verb_lemmas <- udpipe(verbs$token, object = udpipe_load_model("/home/knut/Documents/english-ewt-ud-2.5-191206.udpipe")) %>% distinct.(lemma) %>% filter.(nchar(lemma)>2)


verb_lemmas

## # A tidytable: 13 × 1
##    lemma    
##    <chr>    
##  1 use      
##  2 document 
##  3 building 
##  4 generate 
##  5 develop  
##  6 embed    
##  7 track    
##  8 associate
##  9 look     
## 10 designe  
## 11 add      
## 12 prepare  
## 13 work

Create cross products from atomic query units:

short_tails <- c("cluster documents", "word2vec", "long former model", "bag of words", "text analysis", "sentence embedding", "sentence transformer", "word embeddings", "document embeddings", "paragraph embeddings", "word vectors", "paragraph vectors", "doc2vec", "word mover distance", "topic model", "embedding sentences", "document feature vector")

short_tail_elements_beginning <- c("sentence", "word", "document", "paragraph", "word2vec", "doc2vec")
bigram_ending <- c("embedding", "embeddings", "vector","vectors", "feature vector", "feature fectors", "topic model", "transformer", "transformers")

short_tails_all <- crossing(short_tail_elements_beginning, bigram_ending) 

short_tails_all <- paste(short_tails_all %>% pull(1), short_tails_all %>% pull(2)) %>% c(short_tails) %>% unique()



basic_long_tails_lemmas <- crossing(verb_lemmas$lemma, short_tails_all)
basic_long_tails_tokens <- crossing(verbs$token, short_tails_all)


basic_long_tails <- paste(basic_long_tails_lemmas %>% pull(1), basic_long_tails_lemmas %>% pull(2)) %>% unique() %>% 
  c(
    (paste(basic_long_tails_tokens %>% pull(1), basic_long_tails_tokens %>% pull(2)) %>% unique())
  ) %>% unique()



qs <- c("how to", "what is", "what are")





question_queries <- paste(qs, basic_long_tails) %>% unique() %>% c(
  paste(qs, short_tails_all) %>% unique()
)



question_queries %>% as.data.frame() %>% sample_n(30) %>% kableExtra::kable()

.
what are look sentence embedding
how to paragraph embedding
what is embed word feature vector
what are embedding paragraph embedding
what are building document feature fectors
what are ’s sentence transformers
what are tracking sentence transformers
how to associate word2vec transformers
what are sentence feature fectors
what are tracking doc2vec
what is looking word feature vector
how to embed word2vec transformers
what are designe paragraph feature vector
how to add doc2vec embedding
how to looking word2vec transformers
what is tracking paragraph transformer
what is paragraph embeddings
how to embedding paragraph embeddings
how to embedding doc2vec embedding
what is work paragraph feature fectors
what are associated document transformer
what are associate doc2vec transformer
what is ’s doc2vec vector
how to associated paragraph vector
what are building word2vec vectors
what are work sentence feature vector
how to is doc2vec feature vector
what are preparing text analysis
what are add document feature fectors
what is is doc2vec topic model

Finally we add realted searches on the short-tail queries. Then we will download the search volumes for all candidate queries also from former phrase extraction techniques. I use keywords everywhere for convenience.

library(googleSuggestQueriesR)

# this takes a while
# keyword_suggestions <- googleSuggestQueriesR::suggest_keywords(
#   queries =short_tails_all,
#   lang = "en",
#   interval = 3, enhanced = F)




# all_queries <- c(short_tails_all, basic_long_tails, question_queries, questions, all_verbphrases %>% pull(1), similar_ngrams_ft %>% rownames(), similarities %>% pull(ngram), keyword_suggestions) %>% unique()
# 
# 
# 
# library(kwewr)
# load("/home/knut/Documents/kwapikey.RData")
# 
# 
# 
# batches <- split(all_queries, rep_len(1:100, length(all_queries)))

get_volumes <- function(kws){
  kwewr::get_keyword_data(keywords =  kws, country = "us", api_key = kwapikey)
} 


# avoiding costs of api on knitr, leaving the code here 

#vols <- purrr::map(batches, purrrogress::with_progress(get_volumes))
#save(vols, file = "content/blog/2022-04-17-how-to-scrape-google-search-results-using-r-step-by-step-guide/doc_embs_volumes2.Rdata")


load("~/Desktop/website/content/blog/2022-04-17-how-to-scrape-google-search-results-using-r-step-by-step-guide/doc_embs_volumes2.Rdata")


vols_df <- vols$`1`$data

for (i in 2:length(vols)){
  vols_df <- bind_rows.(vols_df, vols[[i]]$data)
  
}





volumes <- vols_df %>% arrange(desc(vol))

volumes %>% head(300) %>% select(keyword, vol) %>% kableExtra::kable()

keyword	vol
machine learning	49500
how to get	49500
conflicts of interest	27100
hugging face	9900
you are here	8100
word2vec	6600
text analysis	4400
what is semantic	4400
topic modeling	2900
topic model	2900
word embedding	2400
word embeddings	2400
bag of words	1900
supervised machine learning	1000
doc2vec	1000
sentence transformer	880
sentence transformers	880
text analysis example	590
sentence embedding	590
word vector	590
word vectors	590
embedding sentences	590
work with us	590
word2vec python	590
high dimensional data	480
bag of words model	480
topic modeling python	480
what is text analysis	480
generate document	480
develop word	480
get word	390
explain about transformer	390
data availability statement	390
text analysis python	390
text analysis tool	390
bag of words python	320
text analysis software	320
text analysis in r	320
word2vec explained	260
importance of sentences	210
how to develop	210
word2vec gensim	210
topic modelling nlp	170
topic modeling lda	170
how to get word	170
analysis in r	170
transformer examples	140
transformers examples	140
a novel approach	140
document vector	140
cluster documents	140
word mover distance	140
to get a	140
word embeddings python	140
bert word embeddings	140
based language	140
doc2vec gensim	110
machine learning for text analysis	110
what is word2vec	110
word2vec embeddings	110
word2vec paper	110
word2vec online	110
document embedding	90
a guide to	90
text analysis online	90
what is word embeddings	90
word2vec examples	90
machine learning for text	90
word embedding example	90
word embeddings example	90
word2vec tutorial	90
subscribe to blog	70
topic modeling with bert	70
topic modeling algorithms	70
embedded sentences examples	70
what is semantic scholar	70
word2vec model	70
bag of words nlp	50
deep learning based	50
text analysis structure	50
document vectors	50
doc2vec vs word2vec	50
topic modeling r	50
topic modeling example	50
sentence transformers github	50
word embedding word2vec	50
word embeddings word2vec	50
supervised topic model	50
word embeddings visualization	50
word2vec vs bert	40
bag of words sklearn	40
bag of words sentiment analysis	40
transformers word search	40
word transformer	40
word transformers	40
topic modelling techniques	40
bag of words model python	30
doc2vec explained	30
sentence embeddings	30
topic modeling deep learning	30
document management system features	30
using partition	30
learning based	30
word embedding techniques	30
document vector icon	30
of the law	30
text analysis examples pdf	20
transformers word logo	20
doc2vec paper	20
paragraph vector	20
sentence embedding bert	20
paragraph vectors	20
word vector nlp	20
sentence embeddings bert	20
word2vec embedding	20
word embeddings tutorial	20
how to cluster	20
word embedding bert	20
word embedding pytorch	20
word embedding algorithms	20
word embedding tensorflow	20
transformers pipeline example	10
word embeddings in python	10
bag of words paper	10
word transformer majuscule en minuscule	10
embedding word2vec	10
paragraph vector python	10
word transformer tableau en texte	10
text analysis english	10
create word embeddings	10
document transformer	10
doc2vec topic modeling	10
transformers word search printable	10
associated with this	10
doc2vec pretrained model	10
doc2vec gensim tutorial	10
word vector graphics	10
doc2vec github	10
word mover distance python	10
word vector logo	10
guide to building	10
word vectorization techniques	10
word vectorization python	10
what is word vector	10
sentence embedding clustering	10
what are word vectors	10
word vectors in nlp	10
the geometry of	10
word vectors online	10
word vectors download	10
word2vec embeddings python	10
embedded sentences exercises	10
cluster word embeddings	10
embedded sentences in english	10
document topic modeling	10
embedded sentences exercises with answers	10
word2vec vector	10
embedded sentences grammar	10
word2vec vectors	10
word2vec topic modeling	10
transformer document word en pdf	10
using word2vec	10
document level sentiment classification	10
transformer document pdf en word gratuit	10
word2vec methods	10
new transformers names	10
document vector nlp	10
computation and language	10
document vector png	10
with deep learning	10
arabic sentiment analysis	10
learning word embeddings	10
document vector representation	10
word embedding similarity	10
to get word	10
learn word2vec	10
machine learning for	10
document folder vector	10
word embeddings paper	10
document embedding with paragraph vectors	10
word embeddings in r	10
code and data	10
full text available	10
embedding topic model	10
word embedding topic model	10
doc2vec embedding	0
add sentence transformer	0
apply doc2vec vectors	0
associate word topic model	0
base long former model	0
building word2vec transformer	0
create sentence embeddings	0
develop doc2vec topic model	0
document word feature fectors	0
generate document vector	0
get word2vec feature fectors	0
improve paragraph vector	0
link doc2vec embeddings	0
map text analysis	0
motivating document feature vector	0
recommender word vectors	0
share paragraph feature vector	0
supervise bag of words	0
use sentence transformer	0
’s doc2vec vectors	0
added word topic model	0
are long former model	0
associated word2vec transformer	0
based sentence embeddings	0
designing doc2vec topic model	0
embedding word embeddings	0
hugging document transformers	0
improving word2vec embeddings	0
k paragraph transformers	0
linking doc2vec embedding	0
mapping sentence vectors	0
shared document feature fectors	0
supervised word vector	0
what is add paragraph feature fectors	0
what are answer word2vec vectors	0
how to associate sentence topic model	0
what is base doc2vec vector	0
what are building word mover distance	0
how to create embedding sentences	0
what is designe word2vec topic model	0
what are document sentence embedding	0
how to generate doc2vec feature vector	0
what is get word embedding	0
what are improve document transformer	0
how to learn word2vec embedding	0
what is map paragraph transformer	0
what are motivating doc2vec	0
how to recommender sentence vector	0
what is share document embeddings	0
what are subscribe word transformers	0
how to use paragraph embeddings	0
what is work word2vec vector	0
what are added sentence feature vector	0
how to are doc2vec transformers	0
what is associated word feature vector	0
what are based document vectors	0
how to creating word2vec feature vector	0
what is embedding paragraph vectors	0
what are hugging doc2vec feature fectors	0
how to improving topic model	0
what is k document topic model	0
what are learning word2vec	0
how to mapping paragraph topic model	0
what is shared cluster documents	0
what are supervised sentence transformers	0
how to document feature vector	0
using document	0
word embedding to documents embedding without	0
model for document	0
for background linking of news	0
and search tools	0
to get a feature	0
using partition averaging on bag of	0
what is paragraph and essay	0
word network topic model python	0
doc2vec embeddings	0
add sentence transformers	0
apply document embedding	0
associate word transformer	0
base paragraph embedding	0
building word2vec transformers	0
create sentence feature fectors	0
develop doc2vec transformer	0
document word feature vector	0
generate document vectors	0
get word2vec feature vector	0
improve paragraph vectors	0
link doc2vec feature fectors	0
map topic model	0
motivating document topic model	0
recommender word2vec	0
share paragraph topic model	0
supervise cluster documents	0
use sentence transformers	0
’s document embedding	0
added word transformer	0
are paragraph embedding	0
associated word2vec transformers	0
based sentence feature fectors	0
designing doc2vec transformer	0
embedding word feature fectors	0
hugging document vector	0
improving word2vec feature fectors	0
k paragraph vector	0
linking doc2vec embeddings	0
mapping text analysis	0
shared document feature vector	0
supervised word vectors	0
what are add paragraph feature vector	0
how to apply bag of words	0
what is associate sentence transformer	0
what are base doc2vec vectors	0
how to building word topic model	0
what is create long former model	0
what are designe word2vec transformer	0

Conclusion

We can see a most long-tail query candidates are rejected and there is still some noise in this list. We can scrape more SERPs and further tweak how to retrieve related queries. A disadvantage of this (very targeted) approach is, it hits the api a lot, we should add similar keywords from the official api, too for more results. The “how to train document embeddings” niche appears to be rather thin, though: quite a few of the automated keyword suggestions are reasonable keywords people might use. “how to use / train doc2vec” has 0 volume. “how to train word2vec” just started to get minimal volume half a year ago! I might also test this with another api. On other data, I found almost no differences between dataforseo and this api. Dataforseo certainly queries the adwords api if they do not have the keyword stored. We should use broader short tails, such as text analysis, topic modelling, text classification etc.