Background: I am new to word2vec. With applying this method, I am trying to form some clusters based on words extracted by word2vec from scientific publications' abstracts. To this end, I have first retrieved sentences from the abstracts via stanfordNLP and put each sentence into a line in a text file. Then the text file required by deeplearning4j word2vec was ready to process (http://deeplearning4j.org/word2vec).

Since the texts come from scientific fields, there are a lot of mathematical terms or brackets. See the sample sentences below:

The meta-analysis showed statistically significant effects of pharmacopuncture compared to conventional treatment = 3.55 , P = .31 , I-2 = 16 % ) . 

90 asymptomatic hypertensive subjects associated with LVH , DM , or RI were randomized to receive D&G herbal capsules 1 gm/day , 2 gm/day , or identical placebo capsules in double-blind and parallel fashion for 12 months . 

After preparing the text file, I have run word2vec as below:

SentenceIterator iter = new LineSentenceIterator(new File(".../filename.txt"));
        iter.setPreProcessor(new SentencePreProcessor() {
            public String preProcess(String sentence) {
                return sentence.toLowerCase();

    // Split on white spaces in the line to get words
    TokenizerFactory t = new DefaultTokenizerFactory();
    t.setTokenPreProcessor(new CommonPreprocessor());

    log.info("Building model....");
    Word2Vec vec = new Word2Vec.Builder()

    log.info("Fitting Word2Vec model....");

    log.info("Writing word vectors to text file....");

    // Write word vectors
    WordVectorSerializer.writeWordVectors(vec, "abs_terms.txt");

This script creates a text file containing many words withe their related vector values in each row as below:

pills -4.559159278869629E-4 0.028691953048110008 0.023867368698120117 ...
tricuspidata -0.00431067543104291 -0.012515762820839882 0.0074045853689312935 ...

As a subsequent step, this text file has been used to form some clusters via k-means in spark. See the code below:

val rawData = sc.textFile("...abs_terms.txt")
val extractedFeatureVector = rawData.map(s => Vectors.dense(s.split(' ').slice(2,101).map(_.toDouble))).cache()

val numberOfClusters = 10
val numberOfInterations = 100

//We use KMeans object provided by MLLib to run
val modell = KMeans.train(extractedFeatureVector, numberOfClusters, numberOfInterations)


//Get cluster index for each buyer Id
val AltCompByCluster = rawData.map {
    (modell.predict(Vectors.dense(row.split(' ').slice(2,101)


Questions: As a result of the latest scala code above, I have retrieved 10 clusters based on the word vectors suggested by word2vec. However, when I have checked my clusters no obvious common words appeared. That is, I could not get reasonable clusters as I expected. Based on this bottleneck of mine I have a few questions:

1. From some tutorials for word2vec I have seen that no data cleaning is made. In other words, prepositions etc. are left in the text. So how should I apply cleaning procedure when applying word2vec?

2. How can I visualize the clustering results in a explanatory way?

3. Can I use word2vec word vectors as input to neural networks? If so which neural network (convolutional, recursive, recurrent) method would be more suitable for my goal?

4. Is word2vec meaningful for my goal?

  • $\begingroup$ plz i search for an explanation of the word vector written in the text abs_terms.txt, what's the meaning of this values exaclty ?? $\endgroup$
    – ghizlane
    May 20, 2017 at 21:42
  • $\begingroup$ @ghizlanehnini abs_terms is the text file containing the extracted words and their vector values. For example I gave two samples from the words I had (e.g. pills, tricuspidata). You see in the code above a feature named layerSize. It tells that a vector with a size of 100 for each extracted word is created. You can see the details here: deeplearning4j.org/word2vec. Since I am also a beginner in this topic I am still trying to figure out. $\endgroup$ May 21, 2017 at 10:26

1 Answer 1


In my experience, the more data you give word2vec, the better it performs. If none of your clusters make sense, I would throw more data at it. I did something similar to what you are doing and trained with 20k sentences and clusters were bad. When I trained with 2 million sentences the clusters were very good. Now, how to cluster millions of vectors is it's own problem, but you can look online about big data clustering algorithms and experiment with different packages and maybe different ideas that you come up with.

As a first order measure, you can train word to vec and then look at the distances between a few hundred similar words like 'genes' and 'chromosomes', 'molecule' and 'ion'. If the distances are low, then it' s time to start thinking about how to automate clustering.

  • 1
    $\begingroup$ "When I trained with 2 million sentences the clusters were very good" -- how did you evaluate the quality of your clusters? $\endgroup$
    – Jon
    Sep 19, 2017 at 16:20
  • $\begingroup$ This project was for work where we were extracting semantic value from a bunch of social media text. We evaluated clusters by eyeballing clusters which had about average size but were tightly packed. Not the most scientific way, but we made what we thought were reliable conclusions based on these clusters. I don't know how you'd evaluate document clustering if not by eye after letting the machine do the work, and it's entirely possible that most of our results were trash and our analysis was totally wrong. We did get some interesting clusters which had a variety of vocab and similar meaning. $\endgroup$ Sep 19, 2017 at 20:33

Not the answer you're looking for? Browse other questions tagged or ask your own question.