Method, apparatus, server, and storage medium for recalling for search

1. A method for recalling for a search, comprising:

acquiring a search term inputted by a user;

calculating a semantic vector of the search term using a pre-trained neural network model; and

recalling, according to a pre-established index, target documents related to the semantic vector of the search term from candidate documents, the index being established based on the semantic vectors of the candidate documents, and the semantic vectors of the candidate documents being calculated using the pre-trained neural network model, wherein the neural network model is trained by:

extracting, from a user log accumulated by a search engine, items displayed to users for each of a plurality of search terms, using the items clicked by the users as a training positive sample, and using the items not clicked by the users as a training negative sample;

extracting, randomly from the user log, a preset number of random items as a random negative sample, wherein the random items are not clicked for any of the search terms;

acquiring a related negative sample from the user log based on a related search term of the each of the plurality of the search terms, wherein the related negative sample is an item displayed to the users for the related search item and not clicked by the users; and

training the neural network model using the training positive sample, the training negative sample, the random negative sample, and the related negative sample.

2. The method according to claim 1 , wherein the acquiring a related negative sample from the user log based on a related search term of the each of the plurality of the search terms comprises:

acquiring token information of the each of the plurality of the search terms;

acquiring the related search term of the each of the plurality of the search terms from a pre-established inverted index based on the token information, wherein the inverted index is the inverted index of the token information and a candidate search term pre-established using the each of the plurality of the search terms and the token information of the each of the plurality of the search terms in the user log; and

acquiring, from the user log, the items not clicked by the users in recalled items for the related search term as the related negative sample.

3. The method according to claim 2 , wherein the acquiring a related negative sample from the user log based on a related search term of the each of the plurality of the search terms further comprises:

de-duplicating the related negative sample against the training positive sample corresponding to the each of the plurality of the search terms.

4. The method according to claim 1 , wherein the recalling, according to a pre-established index, target documents related to the semantic vector of the search term from candidate documents comprises:

calculating similarities between the semantic vector of the search term and the semantic vectors of the candidate documents in the index; and

recalling, according to the index, the candidate documents corresponding to the semantic vectors having the similarities satisfying a preset condition as the target documents.

5. An apparatus for recalling for a search, comprising:

at least one processor; and

a memory storing instructions, the instructions when executed by the at least one processor, cause the at least one processor to perform operations, the operations comprising:

acquiring a search term inputted by a user; calculating a semantic vector of the search term using a pre-trained neural network model; and

recalling, according to a pre-established index, target documents related to the semantic vector of the search term from candidate documents, the index being established based on the semantic vectors of the candidate documents, and the semantic vectors of the candidate documents being calculated using the pre-trained neural network model, wherein the neural network model is trained by:

extracting, from a user log accumulated by a search engine, items displayed to users for each of a plurality of search terms, using the items clicked by the users as a training positive sample, and using the items not clicked by the users as a training negative sample;

extracting, randomly from the user log, a preset number of random items as a random negative sample, wherein the random items are not clicked for any of the search terms;

acquiring a related negative sample from the user log based on a related search term of the each of the plurality of the search terms, wherein the related negative sample is an item displayed to the users for the related search item and not clicked by the users; and

training the neural network model using the training positive sample, the training negative sample, the random negative sample, and the related negative sample.

6. The apparatus according to claim 5 , wherein the third sample extraction unit comprises:

acquiring token information of the each of the plurality of the search terms;

acquiring the related search term of the each of the plurality of the search terms from a pre-established inverted index based on the token information, wherein the inverted index is the inverted index of the token information and a candidate search term pre-established using the each of the plurality of the search terms and the token information of the each of the plurality of the search terms in the user log; and

acquiring, from the user log, the items not clicked by the users in recalled items for the related search term as the related negative sample.

7. The apparatus according to claim 6 , wherein the third sample extraction unit further comprises:

de-duplicating the related negative sample against the training positive sample corresponding to the each of the plurality of the search terms.

8. The apparatus according to claim 5 , wherein the document recalling module comprises:

calculating similarities between the semantic vector of the search term and the semantic vectors of the candidate documents in the index; and

recalling, according to the index, the candidate documents corresponding to the semantic vectors having the similarities satisfying a preset condition as the target documents.

9. A non-transitory computer-readable storage medium storing a computer program, the computer program when executed by one or more processors, causes the one or more processors to perform operations, the operations comprising:

acquiring a search term inputted by a user;

calculating a semantic vector of the search term using a pre-trained neural network model; and

recalling, according to a pre-established index, target documents related to the semantic vector of the search term from candidate documents, the index being established based on the semantic vectors of the candidate documents, and the semantic vectors of the candidate documents being calculated using the pre-trained neural network model, wherein the neural network model is trained by:

extracting, from a user log accumulated by a search engine, items displayed to users for each of a plurality of search terms, using the items clicked by the users as a training positive sample, and using the items not clicked by the users as a training negative sample;

extracting, randomly from the user log, a preset number of random items as a random negative sample, wherein the random items are not clicked for any of the search terms;

acquiring a related negative sample from the user log based on a related search term of the each of the plurality of the search terms, wherein the related negative sample is an item displayed to the users for the related search item and not clicked by the users; and

training the neural network model using the training positive sample, the training negative sample, the random negative sample, and the related negative sample.