• ./templates/and./static: these two dirs contain the html templates, css and javascript files for this web application.
• backend.py: this file contains functions for getting word similarity, finding near words through a word or its vector, and plotting the 2D or 3D figures for embeddings.
• flask_web.py: this file contains the actions to render the html pages.
• ./mplot3d/: this dir contains the dependency for plotting 3D figures.
• ./pics: this dir contains pictures for the Markdown files.
• ./images: this dir contains the temp figure file generated in the project.
• ./Test Cases: this dir contains the test cases for the front and back ends of this project, including unit test and integration test.

Dependencies of this web application are flask, gensim, sklearn, matplotlib, and numpy, which can be installed using pip.

$ pip3 install flask gensim sklearn matplotlib --user

This web application requires the pretrained GloVe Word Embeddings (whose format should be changed to a loadable one by the Gensim model). This preparation can be done using the following command.

$ python3 -m gensim.scripts.glove2word2vec --input glove.6B.50d.txt --output glove.6B.50d.w2v.txt

or you can run the getGloVe.sh shell script instead.

$ ./getGloVe.sh

Run the web app using the following command.

$ python3 flask_web.py

After that, this web application will be available at port 5200.

Find the words that are highly similar with the given word.

Given two keywords with vectors \vec{A} and \vec{B} respectively, the distance between \vec{A} and \vec{B} are split into n parts. Let \overrightarrow{AB} donates \vec{A} - \vec{B}, then the vector of the words between \vec{A} and \vec{B} can be represented as Vector(i) = A + x*C, where x=i/n and 1<=i<n . Then we can use Vector(i)to find its most similar word vector.

After you click the Near_Analysis button, it will jump to the route /wed, where you can get the top-n near words for a given word.

You can input a word at position 1 and choose the top-n number at position 2. After clicking the Find Near button, the following results will be given, where position 3 indicates the list of similar words.

After you click the Between_Analysis button, it will jump to the route /weds, where you can get the top-n near words to each keyword and the most near word between the two keywords.

You can input two keywords at postion 4 and 5 separately, choose the top-n near words at position 6, and specify how many points you want to get at position 7. After clicking the Find Between button, you 'll get the following results, where x-value means the percentage of the distance between word1 and word2.