Consumer insights analysis using word embeddings
In one embodiment, a method includes receiving a request to identify k steps for a particular entity to acquire a target attribute in public sentiments, accessing a table of word vector relationships, looking up an entity word vector corresponding to the entity name and a target attribute word vector corresponding to the n-gram representing the target attribute using the table, determining a directional vector in the d-dimensional embedding space that connects from the entity word vector to the target attribute word vector, identifying k points on the directional vector that evenly split the directional vector into k+1 segments, selecting, for each of the k points, a word vector that is closest to the point, identifying, for each of the k selected word vectors, a corresponding n-gram by looking up the word vector in the table, and sending a response message comprising the k identified n-grams.
1. A method comprising:
by a first computing device in an online social network, receiving, from a second computing device, a request to identify k steps for a particular entity to acquire a target attribute in public sentiments, wherein the request comprising an entity name for the particular entity and an n-gram representing the target attribute;
by the first computing device, accessing a table of word vector relationships, wherein the table of word vector relationships comprises a plurality of unique n-grams and their corresponding word vectors, 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 first computing device, looking up, using the table, an entity word vector corresponding to the entity name and a target attribute word vector corresponding to the n-gram representing the target attribute;
by the first computing device, determining a directional vector in the d-dimensional embedding space that connects from a point represented by the entity word vector to a point represented by the target attribute word vector;
by the first computing device, identifying k points on the directional vector that evenly split the directional vector into k+1 segments;
by the first computing device, selecting, for each of the k points, using the table and based on a similarity metric, a word vector that is closest to the point;
by the first computing device, identifying, for each of the k selected word vectors, a corresponding n-gram by looking up the word vector in the table; and
by the first computing device, sending, to the second computing device, a response message comprising the k identified n-grams.
2. The method of claim 1 , wherein the plurality of unique n-grams in the table are selected from a corpus of text, wherein the corpus of text was collected from a plurality of user-created content objects in the online social network.
3. The method of claim 2 , wherein the table represents results of a word embedding model trained using the corpus of text as training data.
4. The method of claim 3 , wherein the word embedding model is a word2vec model.
5. The method of claim 1 , wherein the looking up the entity word vector and the target attribute word vector comprises looking up the entity name and the n-gram representing the target attribute in the table.
6. The method of claim 1 , wherein the selecting the k word vectors comprises:
computing, for each of the identified k points, a similarity metric for each word vector in the table to the point; and
selecting, for each of the identified k points, a word vector that has a highest similarity metric to the point.
7. The method of claim 6 , wherein the similarity metric is a cosine similarity, a Euclidean distance, or a Jaccard similarity coefficient.
8. The method of claim 1 , wherein the request comprises one or more conditions characterizing an audience.
9. The method of claim 8 , further comprising:
constructing the corpus of text by collecting text content from content objects created by a group of users of the online social network, wherein the group satisfies the one or more conditions;
identifying a list of unique n-grams appearing in the corpus of text;
training a word embedding model using the corpus of text, wherein the word embedding model generates word vectors corresponding to the unique n-grams; and
generating, using the word embedding model, the table based on the list of unique n-grams and their corresponding word vectors.
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, from a second computing device, a request to identify k steps for a particular entity to acquire a target attribute in public sentiments, wherein the request comprising an entity name for the particular entity and an n-gram representing the target attribute;
access a table of word vector relationships, wherein the table of word vector relationships comprises a plurality of unique n-grams and their corresponding word vectors, wherein each of the word vectors represents a semantic context of a corresponding n-gram as a point in a d-dimensional embedding space;
look up, using the table, an entity word vector corresponding to the entity name and a target attribute word vector corresponding to the n-gram representing the target attribute;
determine a directional vector in the d-dimensional embedding space that connects from a point represented by the entity word vector to a point represented by the target attribute word vector;
identify k points on the directional vector that evenly split the directional vector into k+1 segments;
select, for each of the k points, using the table and based on a similarity metric, a word vector that is closest to the point;
identify, for each of the k selected word vectors, a corresponding n-gram by looking up the word vector in the table; and
send, to the second computing device, a response message comprising the k identified n-grams.
12. The media of claim 11 , wherein the plurality of unique n-grams in the table are selected from a corpus of text, wherein the corpus of text was collected from a plurality of user-created content objects in the online social network.
13. The media of claim 12 , wherein the table represents results of a word embedding model trained using the corpus of text as training data.
14. The media of claim 13 , wherein the word embedding model is a word2vec model.
15. The media of claim 11 , wherein the looking up the entity word vector and the target attribute word vector comprises looking up the entity name and the n-gram representing the target attribute in the table.
16. The media of claim 11 , wherein the selecting the k word vectors comprises:
computing, for each of the identified k points, a similarity metric for each word vector in the table to the point; and
selecting, for each of the identified k points, a word vector that has a highest similarity metric to the point.
17. The media of claim 16 , wherein the similarity metric is a cosine similarity, a Euclidean distance, or a Jaccard similarity coefficient.
18. The media of claim 11 , wherein the request comprises one or more conditions characterizing an audience.
19. The media of claim 18 , wherein the software is further operable when executed to:
construct the corpus of text by collecting text content from content objects created by a group of users of the online social network, wherein the group satisfies the one or more conditions;
identify a list of unique n-grams appearing in the corpus of text;
train a word embedding model using the corpus of text, wherein the word embedding model generates word vectors corresponding to the unique n-grams; and
generate, using the word embedding model, the table based on the list of unique n-grams and their corresponding word vectors.
20. 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, from a second computing device, a request to identify k steps for a particular entity to acquire a target attribute in public sentiments, wherein the request comprising an entity name for the particular entity and an n-gram representing the target attribute;
access a table of word vector relationships, wherein the table of word vector relationships comprises a plurality of unique n-grams and their corresponding word vectors, wherein each of the word vectors represents a semantic context of a corresponding n-gram as a point in a d-dimensional embedding space;
look up, using the table, an entity word vector corresponding to the entity name and a target attribute word vector corresponding to the n-gram representing the target attribute;
determine a directional vector in the d-dimensional embedding space that connects from a point represented by the entity word vector to a point represented by the target attribute word vector;
identify k points on the directional vector that evenly split the directional vector into k+1 segments;
select, for each of the k points, using the table and based on a similarity metric, a word vector that is closest to the point;
identify, for each of the k selected word vectors, a corresponding n-gram by looking up the word vector in the table; and
send, to the second computing device, a response message comprising the k identified n-grams.