CN112825076A

CN112825076A - Information recommendation method and device and electronic equipment

Info

Publication number: CN112825076A
Application number: CN201911144558.6A
Authority: CN
Inventors: 吴泳钢; 余浩; 夏丁胤; 康生巧; 朱光楠
Original assignee: Beijing Sogou Technology Development Co Ltd
Current assignee: Beijing Sogou Technology Development Co Ltd
Priority date: 2019-11-20
Filing date: 2019-11-20
Publication date: 2021-05-21
Anticipated expiration: 2039-11-20
Also published as: CN112825076B

Abstract

The embodiment of the invention provides an information recommendation method, an information recommendation device and electronic equipment, wherein the method comprises the following steps: acquiring a search portrait of a user and an information flow portrait, wherein the search portrait is determined according to search behavior data of the user, and the information flow portrait is determined according to consumption data of the user aiming at information flow; recommending information for the user based on the search representation and the information flow representation; compared with the prior art that information recommendation is performed on the user only based on the information flow portraits, portraits for information recommendation performed on the user are richer, and therefore accuracy of information recommendation can be improved.

Description

Information recommendation method and device and electronic equipment

Technical Field

The present invention relates to the field of data processing technologies, and in particular, to an information recommendation method and apparatus, and an electronic device.

Background

With the development of internet technology and terminal technology, most users are gradually used to acquire information (e.g. news) through an information streaming platform (e.g. news platform) in a terminal. The general information flow platform can recommend information for users, so that the users can obtain interested news and user experience is improved.

The information flow platform is used for recommending information based on a user portrait generated by the user behavior in the information flow platform; obviously, the user portrait generated by the information flow platform only can not be described comprehensively, so that the recommendation accuracy is not enough.

Disclosure of Invention

The embodiment of the invention provides an information recommendation method, which aims to improve the accuracy of information recommendation.

Correspondingly, the embodiment of the invention also provides an information recommendation device and electronic equipment, which are used for ensuring the realization and application of the method.

In order to solve the above problem, an embodiment of the present invention discloses an information recommendation method, which specifically includes: acquiring a search portrait of a user and an information flow portrait, wherein the search portrait is determined according to search behavior data of the user, and the information flow portrait is determined according to consumption data of the user aiming at information flow; and recommending information for the user based on the search portrait and the information flow portrait.

Optionally, the obtaining a search representation of a user includes: acquiring a plurality of pieces of search content from the search behavior data, and the page content of a webpage corresponding to a search result clicked by the user for each piece of search content; performing word segmentation processing on each piece of search content respectively to obtain search word segmentation fragments with the sum of M1; respectively carrying out word segmentation processing on each piece of page content to obtain page word segmentation segments with the sum of M2; mining a search representation of the user based on the M1 search participle segments and the M2 page participle segments; wherein M1 and M2 are both positive integers.

Optionally, the mining the search representation of the user based on the M1 search participle segments and the M2 page participle segments comprises: performing explicit mining based on the M1 search participle segments and the M2 page participle segments to determine an explicit search representation of the user; and/or performing implicit mining on the basis of the M1 search participle fragments and the M2 page participle fragments to determine the implicit search portrait of the user.

Optionally, the explicit search representation comprises at least one of: interest categories, interest tags, and interest keywords.

Optionally, the explicitly mining based on the M1 search participle segments and the M2 page participle segments to determine the interest category of the user includes: based on the M1 search participle segments and the M2 page participle segments, performing category classification by using a category mining model to obtain a first classification result, wherein the first classification result comprises: categories and corresponding category probabilities; selecting the first N1 categories with the maximum category probability as interest categories corresponding to the user; wherein N1 is a positive integer.

Optionally, the performing category classification based on the M1 search participle segments and the M2 page participle segments by using a category mining model to obtain a first classification result, including: determining M1 search word vectors corresponding to the M1 search word segmentation segments and M2 page word vectors corresponding to the M2 page word segmentation segments by adopting the category mining model; calculating a first average of the M1 search term vectors and calculating a second average of the M2 page term vectors; and splicing the first average value and the second average value, and determining a first classification result according to the spliced average value.

Optionally, the explicitly mining based on the M1 search participle segments and the M2 page participle segments to determine the interest tag of the user includes: based on the M1 search participle segments and the M2 page participle segments, performing label classification by adopting a label mining model to obtain a second classification result, wherein the second classification result comprises: tags and corresponding tag probabilities; selecting the first N2 labels with the maximum label probability as interest labels corresponding to the user; wherein N2 is a positive integer.

Optionally, the performing, based on the M1 search participle segments and the M2 page participle segments, tag classification by using a tag mining model to obtain a second classification result, including: determining M1 search word vectors corresponding to the M1 search word segmentation segments and M2 page word vectors corresponding to the M2 page word segmentation segments by adopting the label mining model; performing feature extraction on the M1 search word vectors to obtain A1 search feature vectors, and performing feature extraction on the M2 page word vectors to obtain A2 page feature vectors; determining a second classification result according to the A1 search feature vectors and A2 page feature vectors; wherein A1 and A2 are both positive integers.

Optionally, the explicitly mining based on the M1 search participle segments and the M2 page participle segments to determine the interest keyword of the user includes: for each word segmentation segment, respectively calculating the correlation degree of the word segmentation segment and each other word segmentation segment and calculating the sum of the correlation degrees, wherein the word segmentation segment comprises a search word segmentation segment and a page word segmentation segment; selecting the first N3 word segmentation segments with the maximum sum of the correlation degrees as the interest keywords corresponding to the user; wherein N3 is a positive integer.

Optionally, the determining an implicit search representation of the user based on implicit mining of the M1 search participle segments and the M2 page participle segments includes: mining corresponding implicit semantic vectors by adopting a semantic mining model based on the M1 search participle fragments and the M2 page participle fragments; and taking the implicit semantic vector as an implicit search portrait of the user.

Optionally, the recommending information for the user based on the search representation and the information flow representation includes: recalling articles in an information flow database based on the search portrait to obtain a first candidate article; recalling the article in the information flow database based on the information flow image to obtain a second candidate article; and sequencing the first candidate article and the second candidate article, and recommending information for the user according to a sequencing result.

The embodiment of the invention also discloses an information recommendation device, which specifically comprises: the information flow portrait is determined according to consumption data of the user aiming at information flow; and the recommending module is used for recommending information for the user based on the search portrait and the information flow portrait.

Optionally, the obtaining module includes: the content acquisition sub-module is used for acquiring a plurality of pieces of search content from the search behavior data and the page content of the webpage corresponding to the search result clicked by the user aiming at each piece of the search content; the word segmentation processing submodule is used for respectively carrying out word segmentation processing on each piece of search content to obtain search word segmentation fragments with the sum of M1; respectively carrying out word segmentation processing on each piece of page content to obtain page word segmentation segments with the sum of M2; a mining sub-module for mining the search representation of the user based on the M1 search participle segments and the M2 page participle segments; wherein M1 and M2 are both positive integers.

Optionally, the mining submodule includes: an explicit portrait mining unit, configured to perform explicit mining based on the M1 search participle segments and the M2 page participle segments, and determine an explicit search portrait of the user; and the implicit portrait mining unit is used for performing implicit mining on the basis of the M1 search participle fragments and the M2 page participle fragments to determine the implicit search portrait of the user.

Optionally, the explicit sketch mining unit comprises: an interest category determining subunit, configured to perform category classification based on the M1 search participle segments and the M2 page participle segments by using a category mining model, and obtain a first classification result, where the first classification result includes: categories and corresponding category probabilities; selecting the first N1 categories with the maximum category probability as interest categories corresponding to the user; wherein N1 is a positive integer.

Optionally, the interest category determining subunit is configured to determine, by using the category mining model, M1 search term vectors corresponding to the M1 search term segments and M2 page term vectors corresponding to the M2 page term segments; calculating a first average of the M1 search term vectors and calculating a second average of the M2 page term vectors; and splicing the first average value and the second average value, and determining a first classification result according to the spliced average value.

Optionally, the explicit sketch mining unit comprises: an interest tag determining subunit, configured to perform tag classification by using a tag mining model based on the M1 search word segmentation segments and the M2 page word segmentation segments, to obtain a second classification result, where the second classification result includes: tags and corresponding tag probabilities; selecting the first N2 labels with the maximum label probability as interest labels corresponding to the user; wherein N2 is a positive integer.

Optionally, the interest tag determining subunit is configured to determine, by using the tag mining model, M1 search word vectors corresponding to the M1 search word segmentation segments and M2 page word vectors corresponding to the M2 page word segmentation segments; performing feature extraction on the M1 search word vectors to obtain A1 search feature vectors, and performing feature extraction on the M2 page word vectors to obtain A2 page feature vectors; determining a second classification result according to the A1 search feature vectors and A2 page feature vectors; wherein A1 and A2 are both positive integers.

Optionally, the explicit sketch mining unit comprises: the interest keyword determining subunit is used for respectively calculating the relevance of each participle segment and other participle segments and calculating the sum of the relevance, wherein the participle segment comprises a search participle segment and a page participle segment; selecting the first N3 word segmentation segments with the maximum sum of the correlation degrees as the interest keywords corresponding to the user; wherein N3 is a positive integer.

Optionally, the implicit portrait mining unit is specifically configured to mine a corresponding implicit semantic vector by using a semantic mining model based on the M1 search participle segments and the M2 page participle segments; and taking the implicit semantic vector as an implicit search portrait of the user.

Optionally, the recommendation module includes: the first recall submodule is used for recalling the articles in the information flow database based on the search portrait to obtain a first candidate article; the second recall submodule is used for recalling the articles in the information flow database based on the information flow pictures to obtain a second candidate article; and the sequencing recommendation submodule is used for sequencing the first candidate article and the second candidate article and recommending information for the user according to a sequencing result.

The embodiment of the invention also discloses a readable storage medium, and when instructions in the storage medium are executed by a processor of the electronic equipment, the electronic equipment can execute the information recommendation method in any one of the embodiments of the invention.

An embodiment of the present invention also discloses an electronic device, including a memory, and one or more programs, where the one or more programs are stored in the memory, and configured to be executed by one or more processors, and the one or more programs include instructions for: acquiring a search portrait of a user and an information flow portrait, wherein the search portrait is determined according to search behavior data of the user, and the information flow portrait is determined according to consumption data of the user aiming at information flow; and recommending information for the user based on the search portrait and the information flow portrait.

The embodiment of the invention has the following advantages:

in the embodiment of the invention, a search portrait can be determined according to search behavior data of a user, an information flow portrait determined according to consumption data of the user aiming at information flow can be obtained, and information recommendation is carried out on the user based on the search portrait and the information flow portrait; compared with the prior art that information recommendation is performed on the user only based on the information flow portraits, portraits for information recommendation performed on the user are richer, and therefore accuracy of information recommendation can be improved.

Drawings

FIG. 1 is a flow chart of the steps of an embodiment of a method for information recommendation of the present invention;

FIG. 2a is a flowchart illustrating steps of a method for mining user interest categories according to an embodiment of the present invention;

FIG. 2b is a block diagram of a category mining model of the present invention;

FIG. 3a is a flowchart illustrating steps of a method for mining user interest tags according to an embodiment of the present invention;

FIG. 3b is a block diagram of a tag mining model of the present invention;

FIG. 4 is a flowchart illustrating steps of a method for mining keywords of interest to a user according to an embodiment of the present invention;

FIG. 5a is a flowchart illustrating the steps of an embodiment of a method for mining an implicit search image according to the present invention;

FIG. 5b is a block diagram of a double tower structure of a semantic mining model of the present invention;

FIG. 5c is a block diagram of a single tower structure of a semantic mining model of the present invention;

FIG. 6 is a flow chart of the steps of an alternative embodiment of an information recommendation method of the present invention;

FIG. 7 is a block diagram of an embodiment of an information recommendation device according to the present invention;

FIG. 8 is a block diagram of an alternative embodiment of an information recommendation device of the present invention;

FIG. 9 illustrates a block diagram of an electronic device for information recommendation, in accordance with an exemplary embodiment;

fig. 10 is a schematic structural diagram of an electronic device for information recommendation according to another exemplary embodiment of the present invention.

Detailed Description

In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.

One of the core ideas of the embodiment of the invention is that information recommendation is carried out for a user by combining a search image and an information flow image of the user; the method and the device make up the defect that information recommendation accuracy is insufficient due to the fact that information recommendation is performed on the user only on the basis of the information flow sketch in the prior art, and improve the information recommendation accuracy.

Referring to fig. 1, a flowchart illustrating steps of an embodiment of an information recommendation method according to the present invention is shown, which may specifically include the following steps:

102, obtaining a search portrait of a user and an information flow portrait, wherein the search portrait is determined according to search behavior data of the user, and the information flow portrait is determined according to consumption data of the user aiming at information flow.

And 104, recommending information for the user based on the search portrait and the information flow portrait.

In the embodiment of the invention, when a user enters the information flow platform or performs refreshing operation in the information flow platform, the information flow platform can recommend information for the user, such as news recommendation and topic recommendation in various fields.

In order to improve the accuracy of information recommendation, the embodiment of the invention can enrich the portrait of information recommendation for users. In an implementation manner of the embodiment of the present invention, the portrait of the user in other platforms may be obtained, and then the portrait of the user in other platforms and the portrait in the information flow platform (which may be referred to as an information flow portrait in the following) are used as portraits for information recommendation for the user; thereby realizing the purpose of enriching the portrait of information recommendation for users. In an example of the embodiment of the present invention, the information flow platform may obtain a search portrait of the user in a search platform; information recommendation is then jointly made for the user based on the search representation and the information flow representation.

In the embodiment of the invention, the search image can be determined according to the corresponding search behavior data of the user in the search platform. The search platform and the information flow platform can be independent from each other, and can be two platforms belonging to the same system; of course, the search platform may also be integrated into the information flow platform, and the embodiment of the present invention is not limited thereto. The search behavior data may include multiple types, such as search content (which may refer to information input by a user in a search box), page content (such as a title and a text) of a webpage corresponding to a search result clicked by the user, behavior data (such as sharing, copying, commenting, forwarding, browsing duration, and the like) in the webpage corresponding to the search result, and the like, which are not limited in this embodiment of the present invention.

In the embodiment of the invention, the information flow picture can be determined according to the consumption data of the information flow in the information flow platform by the user. The consumption data for the information flow may include various behavior data of the user in the information flow platform, and may include multiple types, such as page content of a webpage corresponding to a clicked article, behavior data (such as sharing, copying, commenting, forwarding, browsing duration, and the like) in a webpage corresponding to clicked recommendation information, and the like.

In the embodiment of the invention, the information flow platform can collect and structure the page content of the related webpage in advance to obtain the corresponding article; the page content of a web page may correspond to an article. Then storing the collected articles in an information flow database; and recalling articles in an information flow database based on the search portrait and the information flow portrait; and then recommending based on the recalled articles, thereby improving the accuracy of information recommendation.

In summary, in the embodiment of the present invention, a search representation determined according to search behavior data of a user and an information flow representation determined according to consumption data of the user for an information flow may be obtained, and then information recommendation may be performed for the user based on the search representation and the information flow representation; compared with the prior art that information recommendation is performed on the user only based on the information flow portraits, portraits for information recommendation performed on the user are richer, and therefore accuracy of information recommendation can be improved.

How to acquire a search image of a user will be described below.

In the embodiment of the present invention, a method for obtaining a search portrait of a user may be: acquiring a plurality of pieces of search content from the search behavior data, and the page content of a webpage corresponding to a search result clicked by the user for each piece of search content; and mining the search portrait of the user based on a plurality of pieces of search content and corresponding page content.

The search behavior data of the user in the first preset time period may be obtained, and then a plurality of pieces of search content searched by the user may be obtained from the search behavior data of the user. The search content may include words, phrases, and sentences, which is not limited in this embodiment of the present invention. The first preset time period may be set as required, for example, in the last half year, and the embodiment of the present invention is not limited thereto. Then, for each piece of search content, the search result clicked by the user in the search result page corresponding to the piece of search content and the behavior data of the user in the web page corresponding to each clicked search result can be obtained. And determining the page content of the webpage corresponding to the search result clicked by the user aiming at each piece of search content based on the behavior data of the user in the webpage corresponding to each clicked search result. For example, when the browsing duration of the web page corresponding to the clicked search result of the user is greater than a preset threshold, the page content of the web page corresponding to the clicked search result may be determined as the page content of the web page corresponding to the search result clicked by the user for each piece of search content. The preset threshold may be set as required, which is not limited in this embodiment of the present invention. For another example, when the user performs the sharing operation in the web page corresponding to the clicked search result, the page content of the web page corresponding to the clicked search result may also be determined as the page content of the web page corresponding to the search result clicked by the user for each piece of search content; and so on. A search representation of the user may then be mined based on the retrieved plurality of search content and corresponding page content.

In an embodiment of the present invention, an implementation manner of mining a search portrait of the user based on a plurality of pieces of the search content and corresponding page content may be: performing word segmentation processing on each piece of search content respectively to obtain search word segmentation fragments with the sum of M1; respectively carrying out word segmentation processing on each piece of page content to obtain page word segmentation segments with the sum of M2; mining a search representation of the user based on the M1 search participle segments and the M2 page participle segments; wherein M1 and M2 are both positive integers.

The method comprises the following steps of performing word segmentation processing on each piece of search content to obtain corresponding search word segmentation segments; wherein, the sum of all search content corresponding to the search participle segment is M1. The word segmentation processing can be respectively carried out on each page content to obtain corresponding page word segmentation segments; the sum of the page participle segments corresponding to all the page contents is M2. Wherein, M1 and M2 are both positive integers, and the invention does not limit the sizes of M1 and M2.

A search representation of the user may then be mined based on the M1 search participle segments and the M2 page participle segments. One of the ways may be: explicit mining is performed based on the M1 search participle segments and the M2 page participle segments to determine an explicit search representation of the user. The explicit mining may refer to mining explicit semantics, which may refer to semantics understandable by a user, such as expressing semantics in an expression manner of words, sentences, and the like.

In one example of the present invention, the explicit mining may include mining at least one of: category, label, keyword; of course, other information may be mined, and the embodiment of the present invention is not limited thereto. The categories correspond to the categories of articles in the information flow platform, such as sports, movies, news, and the like. The tags may be more accurate and more specific information than the categories described above, and may summarize information of the page content, e.g., the category is sports, the tags may be basketball, football, world cup, super school, etc.; the tags may correspond to tags of articles in an information flow platform. The keywords can be more accurate and more specific information than the tags, and have high relevance with other words in the page content; for example, the category is sports, the label is basketball, and the keyword may be Yao, Yi, etc. Then, the category (subsequently called interest category) possibly interested by the user can be determined from the mined categories; determining tags that are likely to be of interest to the user (subsequently referred to as interest tags) from the mined tags; and determining keywords (subsequently referred to as interest keywords) that may be of interest to the user from the mined keywords. At least one of which may then be selected as an explicit search representation of the user.

The following describes ways of determining the interest category, the interest tag, and the interest keyword of the user, respectively.

Referring to fig. 2a, a flowchart illustrating steps of an embodiment of a method for mining user interest categories according to the present invention is shown.

Wherein, explicitly mining based on the plurality of search contents and the corresponding page contents to determine the interest category of the user, may include the following steps:

step 202, based on the M1 search participle segments and the M2 page participle segments, performing category classification by using a category mining model to obtain a first classification result, where the first classification result includes: categories and corresponding category probabilities.

And 204, selecting the top N1 categories with the maximum category probability as the interest categories corresponding to the user.

In the embodiment of the invention, the M1 search participle segments and the M2 page participle segments can be input into a category mining model; and carrying out category classification on the input word segmentation segments by the category mining model, and outputting a corresponding first classification result. Wherein the first classification result comprises a class and a corresponding class probability; the categories in the first classification result may then be sorted, such as sorted in descending order, according to the category probabilities. Then, the first N1 categories with the maximum category probability are selected as the interest categories corresponding to the user; the N1 is a positive integer, which may be specifically set as required, and the embodiment of the present invention is not limited thereto. For example, the first classification result includes: sports (0.91), movie (0.21), fashion (0.62), life (0.11), and caricatures (0.87)); if N1 is 2, two categories of sports and fashion may be selected as the interest categories of the user.

The category mining model may include multiple types, for example, a depth model such as a FastText model, a bayes (bayes) model, and the like, which is not limited in this embodiment of the present invention. The following describes the above step 202, taking the category mining model as the FastText model shown in fig. 2b as an example. The above step 202 may comprise the following sub-steps:

sub-step 202-2, determining M1 search word vectors corresponding to the M1 search participle fragments and M2 page word vectors corresponding to the M2 page participle fragments by adopting the category mining model;

sub-step 202-4, calculating a first mean of the M1 search term vectors, and calculating a second mean of the M2 page term vectors;

substep 202-6, concatenating the first average value and the second average value, and determining a first classification result according to the concatenated average value.

In the embodiment of the invention, the FastText model can respectively convert M1 search participle segments into corresponding M1 search word vectors and can respectively convert M2 page participle segments into corresponding M2 page word vectors. Then respectively calculating a first average value of the M1 search term vectors and a second average value of the M2 page term vectors; and splicing the first average value and the second average value, and determining a first classification result according to the spliced average value. The spliced average value can be input into a full connection layer for processing, the processing result is input into softmax by the full connection layer for processing, and the softmax outputs a first classification result.

Of course, when the category mining model is other models, M1 search participle segments and M2 page participle segments may be input into other models for processing; wherein, the processing procedures of the M1 search participle segments and the M2 page participle segments are not limited by other models.

In an alternative embodiment of the present invention, a plurality of different category mining models may be used to process the M1 search participle segments and the M2 page participle segments to obtain a plurality of first classification results. Then, the first classification results output by the mining models of various types can be weighted to obtain the final first classification result. Wherein, the weight corresponding to each category mining model can be preset; and for each category, carrying out weighted calculation on the category probability of each category in each first classification result according to the corresponding weight to obtain the final category probability corresponding to the category.

In summary, in the embodiment of the present invention, based on the M1 search participle segments and the M2 page participle segments, a category mining model may be used to perform category classification, so as to obtain a first classification result, where the first classification result includes: categories and corresponding category probabilities; then, selecting the first N1 categories with the maximum category probability as interest categories corresponding to the user; and the accuracy of determining the interest categories of the users can be improved, so that the accuracy of information recommendation is further improved.

Referring to fig. 3a, a flowchart illustrating steps of an embodiment of a method for mining user interest tags is shown.

Wherein, explicitly mining based on the plurality of search contents and corresponding page contents to determine the interest tag of the user, may include the following steps:

step 302, based on the M1 search participle segments and the M2 page participle segments, performing label classification by using a label mining model to obtain a second classification result, where the second classification result includes: tags and corresponding tag probabilities;

and 304, selecting the first N2 labels with the maximum label probability as interest labels corresponding to the user.

In the embodiment of the invention, the M1 search participle segments and the M2 page participle segments can be input into a label mining model; and performing label classification on the input word segmentation segments by using the label mining model, and outputting corresponding second classification results. Wherein the second classification result comprises a label and a corresponding label probability; the labels in the first classification result may then be sorted, such as sorted in descending order, according to the label probabilities. Then, the first N2 labels with the maximum label probability are selected as interest labels corresponding to the user; the N2 is a positive integer, which may be specifically set as required, and the embodiment of the present invention is not limited thereto. For example, the second classification result includes: basketball (0.91), football (0.88), volleyball (0.33) and table tennis (0.11); if N2 is 2, two tags, basketball and football, may be selected as the interest tags for the user.

In this embodiment of the present invention, the tag mining model may include multiple types, such as a depth model, which is not limited in this embodiment of the present invention. The following describes the above step 302 by taking the TextCNN depth model shown in fig. 3b as an example. The above step 302 may include the following sub-steps:

in the substep 302-2, the M1 search word vectors corresponding to the M1 search participle segments and the M2 page word vectors corresponding to the M2 page participle segments are determined by using the tag mining model.

Substep 302-4, performing feature extraction on the M1 search term vectors to obtain A1 search feature vectors, and performing feature extraction on the M2 page term vectors to obtain A2 page feature vectors;

substep 302-6 determines a second classification result from the A1 search feature vectors and the A2 page feature vectors.

In the embodiment of the invention, the TextCNN model can respectively convert M1 search participle segments into corresponding M1 search word vectors, and can respectively convert M2 page participle segments into corresponding M2 page word vectors. The TextCNN model and the FastText model may be different in a manner of converting a word segmentation into a corresponding word vector, and this is not limited in the embodiment of the present invention.

Feature extraction may then be performed on the M1 search term vectors and the M2 page term vectors, respectively, which will be described below by taking feature extraction on the M1 search term vectors as an example. The feature extraction of the M1 search term vectors can be performed for W times, and the processing procedures of the M1 search term vectors each time are as follows in sequence: convolution-activation-batch normalization. W is a positive integer such as 2, which is not limited in this embodiment of the present invention. When convolving M1 search word vectors, it may use multi-scale convolution kernels (e.g., 1-4, 2 convolution kernels are shown in fig. 3 b) to perform convolution to extract semantic features of different scales. After W times of feature extraction, pooling can be performed on the W times of feature extraction results to obtain a1 search feature vectors corresponding to the M1 search word vectors. Correspondingly, according to the above manner, a2 page feature vectors corresponding to M2 page word vectors can be obtained. Wherein, the A1 and the A2 are both positive integers, and are related to the convolution kernel adopted in each feature extraction convolution process.

And splicing the A1 search feature vectors and the A2 page feature vectors, and determining a first classification result according to the spliced feature vectors. Wherein the spliced feature vectors may be input to a plurality of fully-connected layers (2 shown in fig. 3 b) for processing; and inputting the processing result into softmax for processing, and inputting a second classification result into the softmax.

To sum up, in the embodiment of the present invention, based on the M1 search participle segments and the M2 page participle segments, a tag mining model is adopted to perform tag classification, so as to obtain a second classification result, where the second classification result includes: tags and corresponding tag probabilities; and then selecting the first N2 labels with the maximum label probability as the interest labels corresponding to the users. And the accuracy of determining the interest tag of the user can be improved, so that the accuracy of information recommendation is further improved.

Referring to fig. 4, a flowchart illustrating steps of an embodiment of a method for mining user interest keywords according to the present invention is shown.

The explicit mining based on the search contents and the corresponding page contents to determine the interest keywords of the user may include the following steps:

step 402, for each participle segment, calculating the correlation degree of the participle segment and each other participle segment and calculating the sum of the correlation degrees, wherein the participle segment comprises a search participle segment and a page participle segment.

And step 404, selecting the first N3 word segmentation segments with the largest sum of the correlation degrees as the interest keywords corresponding to the user.

In the embodiment of the present invention, for each of M1 search participle segments, the (M1-1) correlation degrees between the search participle segment and other (M1-1) search participle segments may be respectively calculated, and the M2 correlation degrees between the search participle segment and each M2 page participle segment may be respectively calculated. The sum of the (M1+ M2-1) correlations for the search participle segment is then calculated. Correspondingly, for each of the M2 paged word segments, (M2-1) degrees of correlation between the paged word segment and other (M2-1) paged word segments can be respectively calculated, and M1 degrees of correlation between the paged word segment and each M1 search word segments can be respectively calculated. Then, the sum of the (M1+ M2-1) correlations corresponding to the word segmentation of the page is calculated. And then selecting the top N3 word segmentation segments with the maximum correlation sum as the interest keywords corresponding to the user. The N3 is a positive integer, which may be specifically set as required, and the embodiment of the present invention is not limited thereto.

In one example of the present invention, a language model such as a Skip-Gram model can be used to calculate the relevance of the participle segment to other participle segments. Wherein the Skip-Gram model is trained based on Negative Sampling, and a random gradient ascent method is used in a gradient iteration process of reverse training.

Wherein, in the training process, the Skip-Gram model inputs: training set (including corpus samples), dimension size Mcount of word vector, context size 2c of Skip-Gram, step length η, number neg of negative samples. And (3) outputting: model parameter theta corresponding to each word in vocabulary, and all word vectors x_w. In one example, the Skip-Gram model training process is as follows: 1. all model parameters θ, all word vectors w are initialized randomly. 2. For each training sample (context (w)₀),w₀) Negative sampling to obtain neg negative example headwords w_iNe, i ═ 1, 2. 3. A gradient ascent iterative process is performed, for each sample (context (w) in the training set₀),w₀,w₁,...w_neg) The following treatment is carried out:

a)for i＝1to 2c:

i)e＝0

ii) for j ═ 0to neg, calculated:

g＝(y_j-f)η

iii) word vector update:

b) if the gradient is converged, finishing the gradient iteration and finishing the algorithm, otherwise returning to the step a to continue the iteration.

In the embodiment of the present invention, the articles in the information flow database may determine the corresponding article types according to the above step 202 plus 204, and determine the corresponding article labels according to the above step 302 plus 304; and the step 402 and 404, determining the corresponding article keywords.

In summary, in the embodiment of the present invention, for each word segmentation segment, the relevance between the word segmentation segment and each other word segmentation segment may be calculated respectively, and the sum of the relevance may be calculated, where the word segmentation segment includes a search word segmentation segment and a page word segmentation segment; and then selecting the first N3 word segmentation segments with the maximum correlation sum as the interest keywords corresponding to the user. And the accuracy of determining the interest keywords of the user can be improved, so that the information recommendation accuracy is further improved.

In an embodiment of the present invention, a manner for mining the search portrait of the user based on the M1 search participle segments and the M2 page participle segments may be: and performing implicit mining on the basis of the M1 search participle fragments and the M2 page participle fragments to determine the implicit search portrait of the user. The implicit mining may refer to mining implicit semantics, and the implicit semantics may refer to semantics represented by symbols, such as vectors.

Referring to fig. 5a, a flow diagram of the steps of an embodiment of a method of mining an implicit search image is shown.

And 502, mining corresponding implicit semantic vectors by adopting a semantic mining model based on the M1 search participle fragments and the M2 page participle fragments.

And step 504, taking the implicit semantic vector as an implicit search portrait of the user.

In the embodiment of the invention, the M1 search participle segments and the M2 page participle segments can be input into a semantic mining model; and mining implicit information by adopting a semantic mining model to determine a corresponding implicit semantic vector. The implicit semantic vector may then be used as an implicit search representation of the user; the implicit semantic vector may be a B-dimensional vector, where B is a positive integer such as 128, which is not limited in this embodiment of the present invention.

In one example of the present invention, the semantic mining model may be a neural network model of convolution depth semantics. The training process of the neural network model of the convolution depth semantics is as follows:

in an alternative embodiment of the present invention, reference may be made to fig. 5b for the trained neural network model of convolution depth semantics as a two-tower structure. The neural network model for training the convolution depth semantics comprises an input layer, a representation layer and a matching layer, wherein the representation layer can comprise a convolution layer, a pooling layer and a full connection layer. X is the search content, Y + is the positive sample, Y-is the negative sample, sim (X, Y +) is the cosine distance of the search content from the positive sample, sim (X, Y-) is the cosine distance of the search content from the negative sample.

In the embodiment of the invention, the training data of the neural network model with the convolution depth semantics can be constructed by adopting keywords and article contents of the article; and then, training the neural network model of the convolution depth semantics by adopting the constructed training data. The article keywords of an article, the article content of the article, and the article contents of a plurality of other articles can be used as a set of training data. Wherein the article content may include one of: article title, article text and article abstract; the article text may be a partial text of an article.

Wherein, a plurality of groups of training data can be input for training each time; the training comprises forward training and reverse training:

forward training: and inputting the article keywords, the article contents of the articles and the article contents of a plurality of other articles in each group of training data into the deep semantic matching model. Wherein, the chapter key words in the set of training data can be used as X to be input into the neural network model of the convolution depth semantics; taking the article content of the article in the training data as Y +, and inputting the article content into a neural network model of convolution depth semantics; and the article contents of a plurality of other articles are respectively used as Y-and input into the neural network model with convolution depth semantics. After the neural network model of the convolution depth semantic outputs the cosine distances corresponding to each group of training data after processing the training data, wherein the cosine distances corresponding to a group of training data can include at least two, one is the cosine distance between the article keyword and the article content of the article, and at least one is the cosine distance between the article keyword and the article content of other articles. The larger the cosine distance is, the closer the vector distance between the vector corresponding to the article keyword and the vector distance of the article content is.

Reverse training: in the embodiment of the present invention, the convolutional deep semantic neural network model may adopt a loss function pair for reverse training, wherein a weight of the deep semantic matching model may be adjusted according to the cosine distances and the loss function, wherein the cosine distances may be substituted into the loss function; and then, according to the loss function substituted into the cosine distance, adjusting the weight of the depth semantic matching model, such as adjusting the weights of a convolution layer, a pooling layer and the like of the depth semantic matching model.

The formula for calculating the cosine distance between the article keyword and the positive sample/negative sample may be as follows:

wherein, X is an article keyword, Y is a positive sample (Y +) or a negative sample (Y-), p is a word vector of one dimension of the article keyword, q is a word vector of the dimension corresponding to the article keyword of the positive sample/the negative sample, and k is the total dimension of the word vector.

Wherein the loss function is as follows:

where Y' includes positive and negative examples and γ is a hyperparameter.

And then mining implicit semantic vectors corresponding to the M1 search participle fragments and the M2 page participle fragments by adopting a trained neural network model of convolution depth semantics.

Wherein, the neural network model for outputting convolution depth semantics of the implicit semantic vector may be a single tower structure, including: input layer and presentation layer, reference may be made to FIG. 5 c.

The M1 search participle segments and the M2 page participle segments can be input into a trained neural network model of convolution depth semantics, and the neural network model of convolution depth semantics can output corresponding implicit semantic vectors. The input layer of the neural network model with convolution depth semantics can convert input data into an input sequence, and then convert the input sequence into a word vector and output the word vector to the presentation layer. The convolutional layer of the presentation layer may perform convolution calculation on the input word vector and then output a convolution result obtained by the convolution calculation to the pooling layer. And the pooling layer performs pooling calculation on the convolution result and then outputs the obtained pooling result to the full-connection layer. And the full connection layer processes the pooling result and outputs an implicit semantic vector.

In the embodiment of the present invention, the article in the information flow database may also determine the corresponding implicit semantic vector according to the

above step

502 and 204.

In an optional embodiment of the present invention, to further improve the accuracy of determining the user search representation, after the search content is obtained from the search behavior data and the page content of the web page corresponding to the search result clicked by the user for the search content: aggregating the search content and the page content into P classes; mining the search representation of the user based on the search content and the page content, wherein the mining comprises the following steps: for each class in the P classes, mining a search portrait corresponding to the class based on search content and page content corresponding to the class; and performing weighted calculation on the search portrait corresponding to each class to determine the search portrait of the user. Wherein P is a positive integer. For example, the user searches 100 times, wherein 10 searches are related to "health preserving", 60 searches are related to "sports", 20 searches are related to "movie & TV" and 10 searches are related to "fashion". Then the 100 searches can be grouped into 4 categories: health preserving, sports and film and television; and each category comprises corresponding search content and page content of a webpage corresponding to a search result clicked by the user aiming at the search content. Meanwhile, the weight corresponding to each class can be determined, and can be the ratio of the searching frequency to the total searching frequency; for example: 0.1 part of health preserving, 0.6 part of sports, 0.2 part of film and television and 0.1 part of fashion.

Then, aiming at each class in the P classes, mining the search portrait corresponding to the class based on the search content and the page content corresponding to the class. And then according to the weight corresponding to each class, carrying out weighted calculation on the search portrait corresponding to each class to determine the search portrait of the user.

In the embodiment of the invention, at least one of the following information flow pictures can be obtained according to the consumption data of the user aiming at the information flow: interest categories, interest tags and interest keywords; the manner of obtaining the information flow representation is similar to the manner of obtaining the user explicit search representation, and is not described herein again. In addition, other information flow images, such as user age, user gender, and the like, may also be obtained according to consumption data of the user for the information flow, which is not limited in this embodiment of the present invention.

In the embodiment of the present invention, the search image and the information stream image may be generated in advance according to the above steps, and are directly obtained when information recommendation needs to be performed for a user.

In summary, in the embodiment of the present invention, based on the M1 search participle segments and the M2 page participle segments, a semantic mining model is adopted to mine corresponding implicit semantic vectors, so as to mine implicit features of user search behavior data; and then, the implicit semantic vector is used as an implicit search image of the user, so that information recommendation can be performed subsequently based on implicit characteristics of user search behavior data, and the accuracy of information recommendation is further improved.

Referring to fig. 6, a flowchart illustrating steps of an alternative embodiment of the information recommendation method of the present invention is shown, which may specifically include the following steps:

step 602, obtaining a search image and an information flow image of a user.

And step 604, recalling the articles in the information flow database based on the search image to obtain a first candidate article.

And step 606, recalling the article in the information flow database based on the information flow image to obtain a second candidate article.

And 608, sorting the first candidate information and the second candidate information, and recommending information for the user according to a sorting result.

In the embodiment of the invention, when information recommendation is required for a user, a search portrait and an information flow portrait of the user can be obtained; articles in an information flow database are then recalled based on the search representation and the information flow representation, respectively.

And each article in the information flow database determines the corresponding article category, article label, article keyword and implicit semantic vector in advance. In one example of the invention, the category of each article in the information flow database can be matched with the interest category in the search portrait, the article label is matched with the interest label in the search portrait, the article keyword is matched with the interest keyword in the search portrait, and the implicit semantic vector of the article is matched with the implicit search portrait; the article matching the search representation is recalled from the information flow database, and subsequently referred to as a first candidate article. In an example of the present invention, the category of each article in the information flow database may be matched with the interest category in the information flow image, the article tag may be matched with the interest tag in the information flow image, and the article keyword may be matched with the interest keyword in the information flow image; recalling the article matched with the information flow portrait from an information flow database. And then, articles which are recalled from the information flow database and matched with the information flow images can be screened according to other information flow images to obtain a second candidate article. Of course, the manner of recalling the article from the information flow database based on the search representation or the information flow representation includes various ways, and the embodiment of the present invention is not limited thereto.

Then, the first candidate article and the second candidate article may be ranked, such as a reverse ranking, and information recommendation may be performed for the user according to a ranking result.

Secondly, for a cold start user of an information flow platform, the image of the information flow is sparse, and the prior art cannot accurately push the information flow; the information recommendation method provided by the embodiment of the invention can recommend information for the cold start user based on the search image of the cold start user; the problem of low recommendation accuracy caused by sparse images of the cold-start user information flow is solved.

Thirdly, information recommendation is carried out on the user based on the search portrait and the information flow portrait, so that more various information can be recommended to the user; the diversity of the recommendation information is enriched. And the information which is read by the information recommending platform but searched by the searching platform can be recommended for the user, so that the user interest can be explored, the information flow picture can be enriched, and the information recommending precision is further improved.

It should be noted that, for simplicity of description, the method embodiments are described as a series of acts or combination of acts, but those skilled in the art will recognize that the present invention is not limited by the illustrated order of acts, as some steps may occur in other orders or concurrently in accordance with the embodiments of the present invention. Further, those skilled in the art will appreciate that the embodiments described in the specification are presently preferred and that no particular act is required to implement the invention.

Referring to fig. 7, a block diagram of an embodiment of an information recommendation apparatus according to the present invention is shown, which may specifically include the following modules:

an obtaining module 702, configured to obtain a search representation and an information stream representation of a user, where the search representation is determined according to search behavior data of the user, and the information stream representation is determined according to consumption data of the user for an information stream;

and a recommending module 704 for recommending information for the user based on the search representation and the information flow representation.

Referring to fig. 8, a block diagram of an alternative embodiment of an information recommendation device of the present invention is shown.

In an optional embodiment of the present invention, the obtaining module 702 includes:

a content obtaining sub-module 7022, configured to obtain multiple pieces of search content from the search behavior data, and page content of a web page corresponding to a search result clicked by the user for each piece of search content;

a word segmentation processing submodule 7024, configured to perform word segmentation processing on each piece of search content, respectively, to obtain search word segmentation segments whose sum is M1; respectively carrying out word segmentation processing on each piece of page content to obtain page word segmentation segments with the sum of M2;

a mining submodule 7026 for mining the search representation of the user based on the M1 search participle segments and the M2 page participle segments; wherein M1 and M2 are both positive integers.

In an alternative embodiment of the present invention, the mining submodule 7026 includes:

an explicit portrait mining unit 70262, configured to perform explicit mining based on the M1 search participle segments and the M2 pageparticiple segments, to determine an explicit search portrait of the user;

an implicit portrait mining unit 70264, configured to perform implicit mining based on the M1 search participle segments and the M2 pageparticiple segments, to determine an implicit search portrait of the user.

In an alternative embodiment of the invention, the explicit search representation comprises at least one of: interest categories, interest tags, and interest keywords.

In an alternative embodiment of the present invention, the explicit image mining unit 70262 includes:

an interest category determining subunit 702622, configured to perform category classification based on the M1 search participle segments and the M2 page participle segments by using a category mining model, and obtain a first classification result, where the first classification result includes: categories and corresponding category probabilities; selecting the first N1 categories with the maximum category probability as interest categories corresponding to the user; wherein N1 is a positive integer.

In an optional embodiment of the present invention, the interest category determining subunit 702622 is configured to determine, by using the category mining model, M1 search term vectors corresponding to the M1 search term segments and M2 page term vectors corresponding to the M2 page term segments; calculating a first average of the M1 search term vectors and calculating a second average of the M2 page term vectors; and splicing the first average value and the second average value, and determining a first classification result according to the spliced average value.

an interest tag determining subunit 702624, configured to perform tag classification by using a tag mining model based on the M1 search participle segments and the M2 page participle segments, to obtain a second classification result, where the second classification result includes: tags and corresponding tag probabilities; selecting the first N2 labels with the maximum label probability as interest labels corresponding to the user; wherein N2 is a positive integer.

In an optional embodiment of the present invention, the interest tag determining subunit 702624 is configured to determine, by using the tag mining model, M1 search term vectors corresponding to the M1 search term segments and M2 page term vectors corresponding to the M2 page term segments; performing feature extraction on the M1 search word vectors to obtain A1 search feature vectors, and performing feature extraction on the M2 page word vectors to obtain A2 page feature vectors; determining a second classification result according to the A1 search feature vectors and A2 page feature vectors; wherein A1 and A2 are both positive integers.

an interest keyword determining subunit 702626, configured to calculate, for each participle segment, a relevance between the participle segment and each other participle segment, and calculate a sum of the relevance, where the participle segment includes a search participle segment and a page participle segment; selecting the first N3 word segmentation segments with the maximum sum of the correlation degrees as the interest keywords corresponding to the user; wherein N3 is a positive integer.

In an optional embodiment of the present invention, the implicit portrait mining unit 70264 is specifically configured to mine a corresponding implicit semantic vector by using a semantic mining model based on the M1 search participles and the M2 page participles; and taking the implicit semantic vector as an implicit search portrait of the user.

In an alternative embodiment of the present invention, the recommending module 704 includes:

a first recall submodule 7042, configured to recall an article in an information flow database based on the search portrait to obtain a first candidate article;

the second recalling submodule 7044 is configured to recall the article in the information flow database based on the information flow image to obtain a second candidate article;

and the ranking recommendation submodule 7046 is configured to rank the first candidate article and the second candidate article, and recommend information to the user according to a ranking result.

For the device embodiment, since it is basically similar to the method embodiment, the description is simple, and for the relevant points, refer to the partial description of the method embodiment.

Fig. 9 is a block diagram illustrating an electronic device 900 for information recommendation, according to an example embodiment. For example, the electronic device 900 may be a mobile phone, a computer, a digital broadcast terminal, a messaging device, a game console, a tablet device, a medical device, an exercise device, a personal digital assistant, and the like.

Referring to fig. 9, electronic device 900 may include one or more of the following components: a processing component 902, a memory 904, a power component 906, a multimedia component 908, an audio component 910, an input/output (I/O) interface 912, a sensor component 914, and a communication component 916.

The processing component 902 generally controls overall operation of the electronic device 900, such as operations associated with display, telephone calls, data communications, camera operations, and recording operations. Processing element 902 may include one or more processors 920 to execute instructions to perform all or a portion of the steps of the methods described above. Further, processing component 902 can include one or more modules that facilitate interaction between processing component 902 and other components. For example, the processing component 902 can include a multimedia module to facilitate interaction between the multimedia component 908 and the processing component 902.

The memory 904 is configured to store various types of data to support operation at the device 900. Examples of such data include instructions for any application or method operating on the electronic device 900, contact data, phonebook data, messages, pictures, videos, and so forth. The memory 904 may be implemented by any type or combination of volatile or non-volatile memory devices such as Static Random Access Memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic or optical disks.

The power component 906 provides power to the various components of the electronic device 900. Power components 906 may include a power management system, one or more power sources, and other components associated with generating, managing, and distributing power for electronic device 900.

The multimedia components 908 include a screen that provides an output interface between the electronic device 900 and a user. In some embodiments, the screen may include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive an input signal from a user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundary of a touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 908 includes a front facing camera and/or a rear facing camera. The front camera and/or the rear camera may receive external multimedia data when the electronic device 900 is in an operating mode, such as a shooting mode or a video mode. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.

The audio component 910 is configured to output and/or input audio signals. For example, the audio component 910 includes a Microphone (MIC) configured to receive external audio signals when the electronic device 900 is in an operational mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signals may further be stored in the memory 904 or transmitted via the communication component 916. In some embodiments, audio component 910 also includes a speaker for outputting audio signals.

I/O interface 912 provides an interface between processing component 902 and peripheral interface modules, which may be keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to: a home button, a volume button, a start button, and a lock button.

The sensor component 914 includes one or more sensors for providing status evaluations of various aspects of the electronic device 900. For example, sensor assembly 914 may detect an open/closed state of device 900, the relative positioning of components, such as a display and keypad of electronic device 900, sensor assembly 914 may also detect a change in the position of electronic device 900 or a component of electronic device 900, the presence or absence of user contact with electronic device 900, orientation or acceleration/deceleration of electronic device 900, and a change in the temperature of electronic device 900. The sensor assembly 914 may include a proximity sensor configured to detect the presence of a nearby object in the absence of any physical contact. The sensor assembly 914 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 914 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

The communication component 916 is configured to facilitate wired or wireless communication between the electronic device 900 and other devices. The electronic device 900 may access a wireless network based on a communication standard, such as WiFi, 2G or 3G, or a combination thereof. In an exemplary embodiment, the communication part 914 receives a broadcast signal or broadcast associated information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communications component 914 further includes a Near Field Communication (NFC) module to facilitate short-range communications. For example, the NFC module may be implemented based on Radio Frequency Identification (RFID) technology, infrared data association (IrDA) technology, Ultra Wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

In an exemplary embodiment, the electronic device 900 may be implemented by one or more Application Specific Integrated Circuits (ASICs), Digital Signal Processors (DSPs), Digital Signal Processing Devices (DSPDs), Programmable Logic Devices (PLDs), Field Programmable Gate Arrays (FPGAs), controllers, micro-controllers, microprocessors or other electronic components for performing the above-described methods.

In an exemplary embodiment, a non-transitory computer readable storage medium comprising instructions, such as the memory 904 comprising instructions, executable by the processor 920 of the electronic device 900 to perform the above-described method is also provided. For example, the non-transitory computer readable storage medium may be a ROM, a Random Access Memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.

A non-transitory computer readable storage medium in which instructions, when executed by a processor of an electronic device, enable the electronic device to perform a method of information recommendation, the method comprising: acquiring a search portrait of a user and an information flow portrait, wherein the search portrait is determined according to search behavior data of the user, and the information flow portrait is determined according to consumption data of the user aiming at information flow; and recommending information for the user based on the search portrait and the information flow portrait.

Fig. 10 is a schematic structural diagram of an electronic device 1000 for information recommendation according to another exemplary embodiment of the present invention. The electronic device 1000 may be a server, which may have large differences due to different configurations or performances, and may include one or more Central Processing Units (CPUs) 1022 (e.g., one or more processors) and a memory 1032, one or more storage media 1030 (e.g., one or more mass storage devices) storing applications 1042 or data 1044. Memory 1032 and storage medium 1030 may be, among other things, transient or persistent storage. The program stored on the storage medium 1030 may include one or more modules (not shown), each of which may include a series of instruction operations for the server. Still further, the central processor 1022 may be disposed in communication with the storage medium 1030, and execute a series of instruction operations in the storage medium 1030 on the server.

The server may also include one or more power supplies 1026, one or more wired or wireless network interfaces 1050, one or more input-output interfaces 1058, one or more keyboards 1056, and/or one or more operating systems 1041, such as Windows Server, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, etc.

An electronic device comprising a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by one or more processors the one or more programs including instructions for: acquiring a search portrait of a user and an information flow portrait, wherein the search portrait is determined according to search behavior data of the user, and the information flow portrait is determined according to consumption data of the user aiming at information flow; and recommending information for the user based on the search portrait and the information flow portrait.

The embodiments in the present specification are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other.

Embodiments of the present invention are described with reference to flowchart illustrations and/or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing terminal to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing terminal to cause a series of operational steps to be performed on the computer or other programmable terminal to produce a computer implemented process such that the instructions which execute on the computer or other programmable terminal provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

While preferred embodiments of the present invention have been described, additional variations and modifications of these embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the embodiments of the invention.

Finally, it should also be noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or terminal. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or terminal that comprises the element.

The information recommendation method, the information recommendation device and the electronic device provided by the invention are described in detail, and specific examples are applied in the text to explain the principle and the implementation of the invention, and the description of the above embodiments is only used to help understanding the method and the core idea of the invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present invention.

Claims

1. An information recommendation method, comprising:

acquiring a search portrait of a user and an information flow portrait, wherein the search portrait is determined according to search behavior data of the user, and the information flow portrait is determined according to consumption data of the user aiming at information flow;

and recommending information for the user based on the search portrait and the information flow portrait.

2. The method of claim 1, wherein obtaining a search representation of a user comprises:

acquiring a plurality of pieces of search content from the search behavior data, and the page content of a webpage corresponding to a search result clicked by the user for each piece of search content;

performing word segmentation processing on each piece of search content respectively to obtain search word segmentation fragments with the sum of M1; respectively carrying out word segmentation processing on each piece of page content to obtain page word segmentation segments with the sum of M2;

mining a search representation of the user based on the M1 search participle segments and the M2 page participle segments; wherein M1 and M2 are both positive integers.

3. The method of claim 2, wherein mining the search representation of the user based on the M1 search participle snippets and M2 page participle snippets comprises:

performing explicit mining based on the M1 search participle segments and the M2 page participle segments to determine an explicit search representation of the user; and/or the presence of a gas in the gas,

and performing implicit mining on the basis of the M1 search participle fragments and the M2 page participle fragments to determine the implicit search portrait of the user.

4. The method of claim 3, wherein the explicit search representation comprises at least one of: interest categories, interest tags, and interest keywords.

5. The method of claim 4, wherein the determining the interest category of the user based on explicit mining of the M1 search participle segments and M2 page participle segments comprises:

based on the M1 search participle segments and the M2 page participle segments, performing category classification by using a category mining model to obtain a first classification result, wherein the first classification result comprises: categories and corresponding category probabilities;

selecting the first N1 categories with the maximum category probability as interest categories corresponding to the user;

wherein N1 is a positive integer.

6. The method of claim 5, wherein the classifying the categories based on the M1 search participle segments and the M2 page participle segments by using a category mining model to obtain a first classification result comprises:

determining M1 search word vectors corresponding to the M1 search word segmentation segments and M2 page word vectors corresponding to the M2 page word segmentation segments by adopting the category mining model;

calculating a first average of the M1 search term vectors and calculating a second average of the M2 page term vectors;

and splicing the first average value and the second average value, and determining a first classification result according to the spliced average value.

7. The method of claim 4, wherein the determining interest tags for the user based on explicit mining of the M1 search participle segments and M2 page participle segments comprises:

based on the M1 search participle segments and the M2 page participle segments, performing label classification by adopting a label mining model to obtain a second classification result, wherein the second classification result comprises: tags and corresponding tag probabilities;

selecting the first N2 labels with the maximum label probability as interest labels corresponding to the user;

wherein N2 is a positive integer.

8. An information recommendation apparatus, comprising:

the information flow portrait is determined according to consumption data of the user aiming at information flow;

and the recommending module is used for recommending information for the user based on the search portrait and the information flow portrait.

9. A readable storage medium, wherein instructions in the storage medium, when executed by a processor of an electronic device, enable the electronic device to perform the information recommendation method of any of method claims 1-7.

10. An electronic device comprising a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by one or more processors the one or more programs including instructions for: