Methods and systems for query segmentation

1. A computer-implemented method of segmenting a search query to identify search results, the method comprising:

receiving a search query comprising a search query string, the search query string including at least a first token and a second token;

identifying the first token is adjacent to the second token in the search query string;

determining a first vector for the first token and a second vector for the second token using a word embeddings model;

determining to include the first and second tokens in a single query segment using a binary classifier to determine a relationship between the first vector and the second vector in vector space by providing a combined vector representation of the first vector and the second vector to the binary classifier, the first and second vector being combined based on the first token being adjacent to the second token in the search query string, wherein the binary classifier is trained from a training set of queries, the queries of the training set comprising annotated query segments determined from ordinal positions of tokens within the queries, and the annotation indicating proper query segmentation; and

processing the single query segment determined by the binary classifier to identify search results from items stored on a database.

2. The method of claim 1 , wherein the combined vector representation is generated by concatenating the first vector determined for the first token and the second vector determined for the second token.

3. The method of claim 1 , further comprising building the word embeddings model based on training data comprising a database of tokens, wherein the database of tokens comprises previous search queries.

4. The method of claim 3 , wherein the word embeddings model is a word2vec or GloVe model.

5. The method of claim 1 , wherein the binary classifier is trained based on vector representations of training search queries comprising segmented search queries having a predetermined level of segmentation accuracy.

6. The method of claim 1 , wherein the binary classifier uses vector representations of only two tokens of the search query string when determining the relationship, the only two tokens comprising the first token and the second token, and the vector representations comprising the first vector and the second vector.

7. The method of claim 1 , further comprising training the classifier using a contiguous bag of words (cbow) or skip-gram architecture.

8. The method of claim 1 , further comprising annotating query segments to generate the annotated query segments for training the binary classifier, wherein the annotations indicate proper query segmentation and the query segments are annotated based on click-through results of previous search queries.

9. An apparatus for segmenting a search query, comprising at least one processor; and one or more —non-transitory computer-readable— media having instructions stored thereon that cause the at least one processor to:

receive a search query comprising a search query string, the search query string including at least a first token and a second token;

identify the first token is adjacent to the second token in the search query string;

determine a first vector for the first token and a second vector for the second token using a word embeddings model;

determine to include the first and second tokens in a single query segment using a binary classifier to determine a relationship between the first vector and the second vector in vector space by providing the first vector and the second vector of the adjacent tokens to the binary classifier, wherein the binary classifier is trained from a training set of queries, the queries of the training set comprising annotated query segments, the annotated query segments determined from ordinal positions of tokens within the queries, wherein the annotations of the annotated query segments indicate proper query segmentation; and

process the single query segment determined by the binary classifier to identify search results from items stored on a database.

10. The apparatus of of claim 9 , wherein the word embeddings model is a word2vec or GloVe model.

11. The apparatus of claim 9 , further comprising padding the first vector and the second vector to generate first and second features respectively, wherein the determination to include the first and second tokens in the single query segment is based on the first and second features.

12. The apparatus of claim 9 , wherein the training set of search queries comprises comprising segmented search queries having a predetermined level of segmentation accuracy.

13. The apparatus of claim 12 , wherein the binary classifier uses vector representations of only two tokens of the search query string when determining the relationship, the only two tokens comprising the first token and the second token, and the vector representations comprising the first vector and the second vector, wherein the first vector and the second vector are combined, and the combined vector is provided to the binary classifier.

14. The apparatus of claim 9 , further comprising:

generating a plurality of query segments based on the first and second tokens; and

identifying, as the single query segment, a dominant query segment of the plurality of query segments, wherein the identified search results are based on searching the database with the dominant query segment.

15. The apparatus of claim 9 , wherein the first and second tokens are identified in the search query string based on one or more token delimiting characters between the first token and the second token.

16. The apparatus of claim 9 , further comprising annotating query segments to generate the annotated query segments for training the binary classifier, wherein the annotations indicate proper query segmentation and the query segments are annotated based on click-through results of the previous search queries.

17. A machine-readable medium comprising instructions that when executed cause one or more hardware processors to perform a method of segmenting a search query, the method comprising:

receiving, by one or more hardware processors, a search query comprising a search query string, the search query string including at least a first token and a second token;

identifying, by the one or more hardware processors, the first token is adjacent to the second token in the search query string;

determining, by the one or more hardware processors, a first vector for the first token and a second vector for the second token using a word embeddings model;

determining to include the first and second tokens in a single query segment using a binary classifier to determine a relationship between the first vector and the second vector in vector space by providing the first vector and the second vector of the adjacent tokens to the binary classifier, wherein the binary classifier is trained from a training set of queries, the queries of the training set comprising annotated query segments determined from ordinal positions of tokens within the queries, and the annotation indicating proper query segmentation; and

processing, by the one or more hardware processors, the single query segment determined by the binary classifier to identify search results from items stored on a database.

18. The machine-readable medium of claim 17 , wherein the method further comprises annotating query segments to generate the annotated query segments for training the binary classifier, wherein the annotations indicate proper query segmentation and the query segments are annotated based on click-through results of the-previous search queries.