Consumer insights analysis using word embeddings

• US 10,685,183 B1
• Filed: 01/04/2018
• Issued: 06/16/2020
• Est. Priority Date: 01/04/2018
• Status: Active Grant

• Current Assignee: Meta Platforms, Inc. (f/k/a Facebook, Inc.)
• Original Assignee: Meta Platforms, Inc. (f/k/a Facebook, Inc.)
• Inventors: Arfa, Jonathan Michael, Nawathe, Nikhil Girish, Kauder, Bryan, Subramanian, Shriram
• Primary Examiner(s): Le, Thuykhanh
• Application Number:
  US15/862,070
• Time in Patent Office: 894 Days
• Field of Search: None
• CPC Class Codes: G06F 40/284 Lexical analysis, e.g. tokenisation or collocates
• CPC Class Codes: G06F 40/295 Named entity recognition
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: H04L 51/214 using selective forwarding
• CPC Class Codes: H04L 51/52 for supporting social networking services

Abstract

In one embodiment, a method includes receiving a request to generate a visualization of public sentiments regarding a particular subject by a plurality of clusters, where each cluster includes a plurality of words semantically close to each other, constructing a first corpus of text by collecting text containing the input n-gram from a plurality of user-created content objects in the online social network, identifying a list of unique n-grams appearing in the first corpus of text, generating a table comprising unique n-grams in the list and their corresponding word vectors using a word embedding model, classifying word vectors in the table into a plurality of clusters based on semantic similarities of the word vectors, and sending, as a response to the request, instructions to display n-grams in the table in a two-dimensional display space, where n-grams corresponding to word vectors that belong to a cluster are displayed together.

Claims

1. A method comprising:

by a computing device in an online social network, receiving a request to generate a visualization of public sentiments regarding a particular subject by a plurality of clusters, wherein each cluster comprises a plurality of words semantically close to each other, wherein the request comprises one or more conditions characterizing a group of users, and wherein the request comprises an input n-gram representing the particular subject;

by the computing device, identifying users of the online social network who satisfy the one or more conditions;

by the computing device, constructing a first corpus of text by collecting text containing the input n-gram from a plurality of content objects in the online social network created by the identified users;

by the computing device, identifying a list of unique n-grams appearing in the first corpus of text;

by the computing device, generating, using a word embedding model, a table comprising unique n-grams in the list and their corresponding word vectors, wherein the word embedding model was trained using a second corpus of text collected from a plurality of user-created content objects in the online social network as training data, wherein each of the word vectors represents a semantic context of a corresponding n-gram as a point in a d-dimensional embedding space;

by the computing device, classifying word vectors in the table into a plurality of clusters based on semantic similarities of the word vectors;

by the computing device, calculating, for each of the plurality of clusters, a Term Frequency-Inverse Document Frequency (TF-IDF) score associated with the cluster;

by the computing device, determining k most relevant clusters to the particular subject based on the calculated TF-IDF scores; and

by the computing device, sending, as a response to the request, instructions to display n-grams associated with the determined k clusters in a two-dimensional display space, wherein n-grams corresponding to word vectors that belong to a cluster are displayed together.

2. The method of claim 1, wherein the word embedding model is a word2vec model.

3. The method of claim 1, wherein the classifying word vectors comprises performing a hierarchical clustering on the word vectors in the table.

4. The method of claim 1, further comprises determining, for each n-gram in the table, a Term Frequency-Inverse Document Frequency (TF-IDF) score.

5. The method of claim 4, wherein the instructions comprise instructions to adjust a font size for an n-gram based at least on a respective TF-IDF score assigned to the n-gram.

6. The method of claim 4, wherein calculating a TF-IDF score associated with a cluster comprises taking an average of determined TD-IDF scores for n-grams corresponding to word vectors that belong to the cluster.

7. The method of claim 4, wherein calculating a TF-IDF score associated with a cluster comprises taking a maximum TD-IDF score among the determined TD-IDF scores for n-grams corresponding to word vectors that belong to the cluster.

8. The method of claim 1, wherein the instructions comprise instructions to assign a font color for n-grams in a semantic cluster.

9. The method of claim 1, wherein

the second corpus of text is constructed by collecting texts from a plurality of content objects in the online social network created by the group of users that satisfy the one or more conditions.

10. The method of claim 9, wherein the content objects were created within a pre-determined period of time.

11. One or more computer-readable non-transitory storage media embodying software that is operable when executed to:

receive a request to generate a visualization of public sentiments regarding a particular subject by a plurality of clusters, wherein each cluster comprises a plurality of words semantically close to each other, wherein the request comprises one or more conditions characterizing a group of users, and wherein the request comprises an input n-gram representing the particular subject;

identify users of the online social network who satisfy the one or more conditions;

construct a first corpus of text by collecting text containing the input n-gram from a plurality of content objects in an online social network created by the identified users;

identify a list of unique n-grams appearing in the first corpus of text;

generate, using a word embedding model, a table comprising unique n-grams in the list and their corresponding word vectors, wherein the word embedding model was trained using a second corpus of text collected from a plurality of user-created content objects in the online social network as training data, wherein each of the word vectors represents a semantic context of a corresponding n-gram as a point in a d-dimensional embedding space;

classify word vectors in the table into a plurality of clusters based on semantic similarities of the word vectors;

calculate, for each of the plurality of clusters, a Term Frequency-Inverse Document Frequency (TF-IDF) score associated with the cluster;

determine k most relevant clusters to the particular subject based on the calculated TF-IDF scores; and

send, as a response to the request, instructions to display n-grams associated with the determined k clusters in a two-dimensional display space, wherein n-grams corresponding to word vectors that belong to a cluster are displayed together.

12. The media of claim 11, wherein the word embedding model is a word2vec model.

13. The media of claim 11, wherein the classifying word vectors comprises performing a hierarchical clustering on the word vectors in the table.

14. The media of claim 11, wherein the software is further operable when executed to determine, for each n-gram in the table, a Term Frequency-Inverse Document Frequency (TF-IDF) score.

15. The media of claim 14, wherein the instructions comprise instructions to adjust a font size for an n-gram based at least on a respective TF-IDF score assigned to the n-gram.

16. The media of claim 11, wherein the instructions comprise instructions to assign a font color for n-grams in a semantic cluster.

17. A system comprising:

one or more processors; and

one or more computer-readable non-transitory storage media coupled to one or more of the processors and comprising instructions operable when executed by one or more of the processors to cause the system to;

receive a request to generate a visualization of public sentiments regarding a particular subject by a plurality of clusters, wherein each cluster comprises a plurality of words semantically close to each other, wherein the request comprises one or more conditions characterizing a group of users, and wherein the request comprises an input n-gram representing the particular subject;

identify users of the online social network who satisfy the one or more conditions;

construct a first corpus of text by collecting text containing the input n-gram from a plurality of content objects in an online social network created by the identified users;

identify a list of unique n-grams appearing in the first corpus of text;

generate, using a word embedding model, a table comprising unique n-grams in the list and their corresponding word vectors, wherein the word embedding model was trained using a second corpus of text collected from a plurality of user-created content objects in the online social network as training data, wherein each of the word vectors represents a semantic context of a corresponding n-gram as a point in a d-dimensional embedding space;

classify word vectors in the table into a plurality of clusters based on semantic similarities of the word vectors;

calculate, for each of the plurality of clusters, a Term Frequency-Inverse Document Frequency (TF-IDF) score associated with the cluster;

determine k most relevant clusters to the particular subject based on the calculated TF-IDF scores; and

send, as a response to the request, instructions to display n-grams associated with the determined k clusters in a two-dimensional display space, wherein n-grams corresponding to word vectors that belong to a cluster are displayed together.