Method of and server for retraining machine learning algorithm

1. A method of retraining a machine learning algorithm (MLA), the MLA having been trained to classify documents based on features thereof, the MLA being executed on a search engine server couplable to a database comprising previously submitted search queries and associated documents presented as search results by the MLA and user interaction data with each of the associated documents, the method being executable by the server, the method comprising:

accessing, by the search engine server via the communication network, a social network resource, the social network resource having a plurality of content items, each content item of the plurality of content items having been submitted by at least one user on the social network resource;

identifying, by the search engine server, at least one content item of the plurality of content items, the at least one content item being associated with a document having been previously presented as a search result in response to a previously submitted search query, the at least one content item meeting a predetermined condition corresponding to:

the at least one content item causing an irregular fluctuation in user interaction data for the document during a predetermined period of time, the irregular fluctuation being determined by a statistical analysis of past user interactions with the document;

acquiring from the database, by the search engine server, a set of document features corresponding to features previously used by the MLA to classify the document;

analyzing, by the search engine server, based on the set of document features of the document, the at least one content item of the plurality of content items to determine if the document has been misclassified by the MLA in response to the search query;

in response to determining that the document has been misclassified:

generating, by the server, based on the document, a training object, the generating including labelling the document with an indication of a misclassification; and

retraining, by the server, the MLA based on the training object.

2. The method of claim 1 , wherein

the method further comprises, prior to the accessing the social network resource:

acquiring, by the server from the database, a plurality of search queries having been previously submitted on the search engine server by a plurality of users, each search query being associated with a respective set of documents having been presented as search results to respective users of the plurality of users;

acquiring, by the search engine server from the database, for each search query of the plurality of search queries, respective traffic information, the respective traffic information including a respective number of submissions; and

determining, by the search engine server, based on the respective number of submissions, the search query.

3. The method of claim 2 , wherein

the respective traffic information further includes a respective traffic source; and wherein

the determining the search query further comprises determining the social network resource based on the respective traffic source.

4. The method of claim 3 , wherein the determining the social network resource is further based on the resource being one of a predetermined list of traffic sources.

5. The method of claim 4 , wherein the determining the search query based on the respective number of submissions is further based on the respective number of submissions being above a predetermined threshold during a predetermined period of time.

6. The method of claim 2 , wherein

the method further comprises, prior to the accessing the social network resource:

acquiring, by the search engine server from the database, a plurality of search queries having been previously submitted on the search engine server by a plurality of users, each search query being associated with a respective set of documents having been presented as search results to respective users of the plurality of users;

acquiring, by the search engine server from the database, for each search query of the plurality of search queries, respective traffic information, the respective traffic information including a respective number of submissions and a respective traffic source; and

determining, by the search engine server, based on the respective number of submissions and the respective traffic source, the resource, the resource being associated with at least one query.

7. The method of claim 1 , wherein

the document is an image; and wherein

the set of document features is a set of image features.

8. The method of claim 1 , wherein the document being misclassified is indicative of the search result being non-responsive to the search query.

9. The method of claim 1 , wherein the document being misclassified is indicative of the search result being non-suitable for the search query.

10. The method of claim 1 , wherein the document being misclassified is indicative of the search result being non-appropriate to a mode of operation of a web browser used by a user submitting the search query.

11. The method of claim 1 , wherein the MLA is a binary classifier.

12. A search engine server for retraining a machine learning algorithm (MLA), the MLA having been trained to classify documents based on features thereof, the server being couplable to a database comprising previously submitted search queries and associated documents presented as search results by the MLA and user interaction data with each of the associated documents, the search engine server being connected to a communication network, the server comprising:

a processor;

a non-transitory computer-readable medium comprising instructions;

the processor, upon executing the instructions, being configured to:

access via the communication network, a social network resource, the social network resource having a plurality of content items, each content item of the plurality of content items having been submitted by at least one user on the social network resource;

identify, at least one content item of the plurality of content items, the at least one content item being associated with a document having been previously presented as a search result in response to a previously submitted search query, the at least one content item meeting a predetermined condition corresponding to:

the at least one content item causing an irregular fluctuation in user interaction data for the document during a predetermined period of time, the irregular fluctuation being determined by a statistical analysis of past user interactions with the document;

acquire from the database, a set of document features corresponding to features previously used by the MLA to classify the document;

analyze, based on a set of document features of the document, the at least one content item of the plurality of content items to determine if the document has been misclassified by the MLA in response to the search query;

in response to determining that the document has been misclassified:

generate, based on the document, a training object, the generating including labelling the document with an indication of a misclassification; and

retrain the MLA based on the training object.

13. The server of claim 12 , wherein

the processor is further configured to, prior to the accessing the resource:

acquire from the database, a plurality of search queries having been previously submitted on the search engine server by a plurality of users, each search query being associated with a respective set of documents having been presented as search results to respective users of the plurality of users;

acquire from the database, for each search query of the plurality of search queries, respective traffic information, the respective traffic information including a respective number of submissions; and

determine, based on the respective number of submissions, the search query.

14. The server of claim 13 , wherein

the respective traffic information further includes a respective traffic source; and wherein

the determining the search query further comprises determining the social network resource based on the respective traffic source.

15. The server of claim 14 , wherein the determining the resource is further based on the resource being one of a predetermined list of traffic sources.

16. The server of claim 15 , wherein the determining the search query based on the respective number of submissions is further based on the respective number of submissions being above a predetermined threshold during a predetermined period of time.

17. The server of claim 12 , wherein

the processor is further configured to, prior to the accessing the resource:

acquire from the database, a plurality of search queries having been previously submitted on the search engine server by a plurality of users, each search query being associated with a respective set of documents having been presented as search results to respective users of the plurality of users;

acquire from the database, for each search query of the plurality of search queries, respective traffic information, the respective traffic information including a respective number of submissions and a respective traffic source; and

determine based on the respective number of submissions and the respective traffic source, the resource, the social network resource being associated with at least one query.

18. The method of claim 1 , wherein the given content item is the at least one content item of the plurality of content items and the condition further comprises:

determining, by the search engine server, the at least one content item of the plurality of content items is associated with a word included within a predetermined word list.