CN112463918B

CN112463918B - Information recommendation method, system, storage medium and terminal equipment

Info

Publication number: CN112463918B
Application number: CN202011091495.5A
Authority: CN
Inventors: 徐东
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2020-10-13
Filing date: 2020-10-13
Publication date: 2023-10-31
Anticipated expiration: 2040-10-13
Also published as: CN112463918A

Abstract

The embodiment of the invention discloses an information recommendation method, an information recommendation system, a storage medium and terminal equipment, which are applied to the technical field of information processing based on artificial intelligence. The information recommendation system adopts a first target association text and a second target association text which are respectively corresponding to the target text and the target word thereof to match the association model, so that a first sample text and a first association text which are matched are obtained, and information recommendation is further carried out based on the first sample text and the first association text. In the process, the text does not need to be subjected to semantic analysis, but the associated text related to the text in terms of semantics, application scenes and the like can be considered, so that the limitation caused in the semantic analysis process can be avoided, more comprehensive text related to the target text, namely the matched first sample and the first associated text, can be provided, and the comprehensive and accurate recommendation of the target text is realized.

Description

Information recommendation method, system, storage medium and terminal equipment

Technical Field

The invention relates to the technical field of information processing based on artificial intelligence, in particular to an information recommendation method, an information recommendation system, a storage medium and terminal equipment.

Background

In the field of information recommendation, a semantic comparison method is generally adopted to conduct recommendation, specifically, after a user inputs search information, a system compares the user search information with information to be recommended stored in the system based on semantics, and information similar to the user search information is recommended to the user.

In the prior art, the semantic information is generally obtained by an artificial intelligence method, for example, a natural language processing (Natural Language Processing, NLP) algorithm, for example, tools such as Ebs or Synonym, is used to obtain the semantic information.

However, in the existing information recommendation method, semantic analysis is required to be performed on the user search information and the information to be recommended respectively to obtain corresponding semantic information, but the semantic analysis process has a certain limit, so that the actual meaning of the user search information and the information to be recommended cannot be comprehensively analyzed, and particularly, the recommendation of small-language information such as Japanese or Korean is difficult due to the fact that the small-language corpus is small and the rule is fuzzy and the understanding of the small language is difficult, so that the existing information recommendation cannot be used for accurately recommending the small-language information.

Disclosure of Invention

The embodiment of the invention provides an information recommendation method, an information recommendation system, a storage medium and terminal equipment, which realize accurate recommendation of target texts.

An aspect of the present invention provides an information recommendation method, including:

acquiring a target text;

word segmentation is carried out on the target text, and target word segmentation corresponding to the target text is obtained;

acquiring a first target association text corresponding to the target text and a second target association text corresponding to the target word;

acquiring an association model, wherein a plurality of first sample texts and characteristic information corresponding to the plurality of first sample texts are stored in the association model, and the characteristic information comprises first association texts;

matching the target text, the obtained target word, the first target associated text and the second target associated text with a first sample text and a first associated text in the associated model respectively to obtain the first sample text and the first associated text matched with the target text, the target word, the first target associated text and the second target associated text;

and recommending information to the target text according to the first sample text and the first associated text obtained by matching.

Another aspect of the embodiment of the present invention provides an information recommendation system, including:

the text acquisition unit is used for acquiring a target text;

the text acquisition unit is further used for word segmentation of the target text to obtain target word segmentation corresponding to the target text, and acquiring a first target association text corresponding to the target text and a second target association text corresponding to the target word segmentation;

the model acquisition unit is used for acquiring an association model, wherein a plurality of first sample texts and characteristic information respectively corresponding to the plurality of first sample texts are stored in the association model, and the characteristic information comprises first association texts;

the matching unit is used for respectively matching the target text, the acquired target word, the first target association text and the second target association text with a first sample text and a first association text in the association model to obtain the first sample text and the first association text which are matched with the target text, the target word, the first target association text and the second target association text;

and the recommending unit is used for recommending information to the target text according to the first sample text and the first associated text which are obtained by matching.

Another aspect of the embodiments of the present invention also provides a computer readable storage medium storing a plurality of computer programs adapted to be loaded by a processor and to perform the information recommendation method according to the one aspect of the embodiments of the present invention.

In another aspect, the embodiment of the invention further provides a terminal device, which comprises a processor and a memory;

the memory is used for storing a plurality of computer programs, and the computer programs are used for being loaded by the processor and executing the information recommendation method according to one aspect of the embodiment of the invention; the processor is configured to implement each of the plurality of computer programs.

It can be seen that, in the method of this embodiment, the information recommendation system matches the association model by using the target text and the first target association text and the second target association text corresponding to the target word respectively, so as to obtain a first sample text and a first association text that are matched, and further, perform information recommendation based on the first sample text and the first association text. In the process, the text does not need to be subjected to semantic analysis, and the associated text related to the text in terms of semantics, application scenes and the like can be considered, so that the limitation caused in the semantic analysis process can be avoided, and more comprehensive text related to the target text, namely the first sample and the first associated text which are matched with each other, can be provided, and the comprehensive and accurate recommendation of the target text is realized.

Drawings

In order to more clearly illustrate the embodiments of the invention or the technical solutions of the prior art, the drawings which are used in the description of the embodiments or the prior art will be briefly described, it being obvious that the drawings in the description below are only some embodiments of the invention, and that other drawings can be obtained according to these drawings without inventive faculty for a person skilled in the art.

Fig. 1 is a schematic diagram of an information recommendation method according to an embodiment of the present invention;

FIG. 2 is a flow chart of a method for recommending information according to an embodiment of the present invention;

FIG. 3 is a flowchart of an information recommendation method according to another embodiment of the present invention;

FIG. 4 is a schematic representation of a text map obtained in another embodiment of the invention;

FIG. 5 is a schematic diagram of an information recommendation method in an embodiment of the invention;

FIG. 6 is a schematic diagram of a user input interface displayed by the information recommendation system in one embodiment of the invention;

FIG. 7 is a schematic diagram of an information recommendation method in another embodiment of the present invention;

FIG. 8a is a schematic diagram of a user input interface displayed by the information recommendation system in another embodiment of the present invention;

FIG. 8b is a schematic diagram of another user input interface displayed by the information recommendation system in another embodiment of the present invention;

FIG. 8c is a schematic representation of a text map in another embodiment of the invention;

FIG. 9 is a schematic diagram of an information recommendation system provided by another embodiment of the present invention;

FIG. 10 is a schematic diagram of an information recommendation system according to an embodiment of the present invention;

fig. 11 is a schematic structural diagram of a terminal device according to an embodiment of the present invention.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

The terms "first," "second," "third," "fourth" and the like in the description and in the claims and in the above drawings, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the invention described herein may be implemented, for example, in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

The embodiment of the invention provides an information recommendation method, which is mainly used for recommending related information according to a target text input by a user, wherein as shown in fig. 1, an information recommendation system can realize information recommendation by the following steps:

acquiring a target text; word segmentation is carried out on the target text, and target word segmentation corresponding to the target text is obtained; acquiring a first target association text corresponding to the target text and a second target association text corresponding to the target word; acquiring an association model, wherein a plurality of first sample texts and characteristic information corresponding to a plurality of first sample texts are stored in the association model, and the characteristic information comprises first association texts; matching the target text, the obtained target word, the first target associated text and the second target associated text with a first sample text and a first associated text in the associated model respectively to obtain the first sample text and the first associated text matched with the target text, the target word, the first target associated text and the second target associated text; and recommending information to the target text according to the first sample text and the first associated text obtained by matching.

In a specific implementation process, the information recommendation system can be applied to terminal equipment, and a user can input a target through an interface provided by the terminal equipment, so that the terminal equipment can recommend related information to a target text in real time according to the method; or in another case, the information recommendation system is applied to the background, when the user inputs the target text through the interface provided by the terminal equipment, the terminal equipment uploads the target text input by the user to the background, and the background recommends relevant information to the target text in real time according to the method. The information recommendation can be mainly applied to recommendation of application programs or recommendation of product information and the like, and can be applied to any application system which can be searched by a user.

Therefore, the text does not need to be subjected to semantic analysis in the process, but the associated text related to the text in terms of semantics, application scenes and the like can be considered, so that the limitation caused in the semantic analysis process can be avoided, more comprehensive text related to the target text, namely the first sample matched with the first associated text, can be provided, and the comprehensive and accurate recommendation of the target text is realized.

An embodiment of the present invention provides an information recommendation method, mainly a method executed by an information recommendation system, where a flowchart is shown in fig. 2, and includes:

step 101, acquiring a target text;

it can be understood that, when a user inputs a text through an interface provided by the information recommendation system, the information recommendation system initiates the information recommendation method flow of the embodiment by taking the text as a target text after acquiring the text input by the user. Or, the user may input a section of voice through an interface provided by the information recommendation system, and after the information recommendation system obtains the section of voice input by the user, the section of voice is converted into text, and the converted text is used as a target text, so as to initiate the information recommendation method flow of the embodiment.

Step 102, word segmentation is carried out on the target text to obtain target word segmentation corresponding to the target text, and a first target association text corresponding to the target text and a second target association text corresponding to the target word segmentation are obtained.

Specifically, the information recommendation system may perform word segmentation on the target text, and may specifically perform word segmentation on the target text by using a word segmentation device, for example, a core word segmentation device or an open source word segmentation device of lucene is used to obtain the target word segmentation. Further, since some target words are nonsensical in terms of semantics, in order to simplify the processing of the target words, the information recommendation system may remove the target words, specifically, may remove the nonsensical target words such as the word gas word and the tense word in the obtained target words, so as to obtain the remaining target words. For example, the target text is "people's name", the corresponding target words are "people", "people" and "names", and the "target word" can be eliminated.

The information recommendation system then acquires a first target associated text of the target text, wherein the first target associated text is text associated with the target text in terms of semantics, application scene and the like, and for example, the associated text of the name of a person can comprise Darcy book and the like; the information recommendation system also obtains a second target associated text of the target word, wherein the second target associated text is text related to the target word in terms of semantics, application scene and the like. For example, the associative text of "applet" may include "WeChat applet" and "qq applet" and the like.

Step 103, obtaining an association model, wherein the association model stores a plurality of first sample texts and characteristic information corresponding to the plurality of first sample texts, and the characteristic information comprises first association texts corresponding to the first sample texts.

The information recommendation system may acquire the association model before initiating the information recommendation method flow of the embodiment, and store the association model in the local storage, and after initiating the information recommendation method flow of the embodiment, the information recommendation system may directly extract the association model from the local storage.

The feature information of each first sample text in the association model refers to feature information of operation behaviors such as searching the first sample text by a user, such as information of heat degree, ranking information, first association text and the like, wherein the heat degree information not only can indicate the number of times the first sample text is searched by the user, but also can indicate whether the occurrence time of the first sample text occurs in a last period of time or not, and can be specifically divided into: high, general, and low, e.g., a certain news occurs within a period of time before the current time and the number of searches or views by the user exceeds a certain number of times, the heat information of the vocabulary related to the news is high; the ranking is mainly used for indicating the times that the first sample text is searched or watched by the user, so that the first sample text with higher ranking is not necessarily higher in heat, but the first sample text with higher heat is also higher in ranking; the first associated text may represent text related to semantic or application scenario aspects of the first sample text, i.e., text that the user easily associates with through the first sample text, such as "multi-click" associated text may include "multi-click mall" and "multi-click" text.

In a specific application example, the obtaining of the association model may be obtained according to a historical search record in the sample recommendation system, where when a user inputs a search text in the sample recommendation system and the sample recommendation system searches according to the search text to obtain a search result, the sample recommendation system may record a piece of search operation information, where the search operation information includes the search text and the corresponding search result, and thus, the search operation information recorded by the sample recommendation system in a period of time is the historical search record.

It should be noted that, the sample recommendation system and the information recommendation system currently executing the information recommendation method may be different systems that are independent of each other, so that the information recommendation system may analyze data in other recommendation systems (referred to as sample recommendation systems in this embodiment), and refer to data in the sample recommendation system to obtain an association model, and store the association model in the current information recommendation system, so that the information recommendation system may perform information recommendation on the obtained target text in real time according to the association model.

Specifically, the information recommendation system may first obtain search operation information in the sample recommendation system, where the search operation information includes a search text and a search result of the sample recommendation system on the search text; acquiring sample word segmentation corresponding to a search text; according to the search results corresponding to the search text, counting the associated texts corresponding to the sample word and the search text respectively, wherein the associated text corresponding to the sample word and the search text respectively is the first associated text stored in the associated model, and the sample word and the search text are the first sample text stored in the associated model.

In general, in the information recommending process, the sample recommending system also performs word segmentation on the search text to obtain sample word segmentation, so that the obtained search result includes a plurality of pieces of first search information obtained based on the whole search text and a plurality of pieces of second search information obtained based on the respective sample word segmentation, and when the information recommending system in this embodiment counts the associated text corresponding to the search text, it is determined that when ranking information and heat information of any piece of first search information meet preset conditions, and when any piece of first search information includes the associated text corresponding to the search text, it is determined that the associated text corresponding to the search text is determined according to any piece of first search information.

The fact that the ranking information and the heat information of any first search information meet preset conditions means that compared with all first search information, ranking and heat of any first search information are higher, for example, ranking is performed in the first N, and heat is performed in the first M, so that the first search information with higher ranking and higher heat or the first search information with higher ranking and general heat can be selected.

In addition, when determining whether any of the first search information includes the associated text corresponding to the search text, the search text may be matched with the title or the main content of the first search information, thereby determining whether the first search information includes the corresponding associated text. For example, if most characters (such as characters greater than P percent, where P is a natural number less than 100, such as 80, etc.) in the search text continuously appear in the first search information, the relevant characters in the first search information are used as associated texts corresponding to the search text, for example, the search text is "applet", a "qq applet" appears in one piece of the first search information, a "WeChat applet" appears in another piece of the first search information, and the "qq applet" and the "WeChat applet" are the associated texts corresponding to the search text.

Similarly, when the information recommendation system counts the associative texts corresponding to the sample words, determining that the ranking information and the heat information of any piece of second search information meet preset conditions, wherein any piece of second search information contains the associative texts corresponding to the sample words, and determining the associative texts corresponding to the sample words according to any piece of second search information.

Further, the information recommendation system may further determine heat information and ranking information corresponding to the sample word segmentation and the search text, where the feature information of the first sample text further includes heat information and ranking information corresponding to the sample word segmentation and the search text, respectively.

For example, the obtained set of search operation information of the sample recommendation system includes a search text a, corresponding sample words are A1 and A2, first search information corresponding to the search text a includes Bi, second search information corresponding to the sample word A1 is Cj, and second search information corresponding to the sample word A2 is Dk, where i, j and k are natural numbers greater than 0. In this way, the first search information with higher ranking and higher heat is determined from the first search information Bi, and then the association text corresponding to the search text A is obtained based on the selected first search information; and obtaining the associated text corresponding to the sample word A1 according to the second search information Cj, and obtaining the associated text corresponding to the sample word A2 according to the second search information Dk.

Step 104, matching the target text, the target word, the first target associated text and the second target associated text with the first sample text and the first associated text in the association model obtained in the step 103, respectively, to obtain a first sample text and a first associated text which are matched with the target text, the target word, the first target associated text and the second target associated text.

Specifically, the target text, the target word, the first target associated text and the second target associated text are respectively matched with each first sample text and the first associated text thereof in the association model, so that a first sample text and a first associated text matched with the target text, the target word, the first target associated text and the second target associated text are obtained.

In this process, the information recommendation system may select, from the above-obtained matched first sample texts and first associated texts, the first sample texts and first associated texts whose ranking information and heat information satisfy preset conditions, such as selecting the first sample texts and first associated texts that are ranked higher and have higher heat (or heat is general), in combination with other feature information of each first sample, such as ranking information and heat information.

And 105, recommending information to the target text according to the first sample text and the first associated text obtained by matching in the step 104.

Specifically, the information recommendation system may directly output the first sample text and the first associated text obtained by matching in the step 104, so as to display the first sample text and the first associated text to the user; alternatively, the information recommendation system may apply the first sample text and the first associated text obtained by the matching to a recommendation process of some specific information, for example, recommendation of a program, recommendation of information such as news or video, and the like.

In addition, it should be noted that, through the steps 101 to 104, a text related to the target text in terms of semantics or application scenario (i.e., the first sample text and the first associated text obtained by matching in the step 104) may be obtained, so that when information recommendation is performed based on the obtained related text, the recommended information is relatively comprehensive, and the probability of accuracy of the recommended information may also be increased. For example, when the target text is small language information, the semantic analysis is limited by specific rules in a small language system, so that the semantic analysis of the small language information is wrong, and the method in the embodiment can avoid the situation.

In order to further improve accuracy of the recommended information, the information recommendation system can obtain semantic feature information corresponding to the search text and the sample word thereof respectively according to the search text, the sample word and a preset semantic determination model in the sample recommendation system, namely, the semantic feature information of the first sample text, and store the semantic feature information into the information recommendation system.

When a user inputs a target text in an information recommendation system in real time, the information recommendation system can acquire semantic feature information corresponding to the target text and the target word according to the target text, the target word and a preset semantic determination model; then matching the determined semantic feature information with the semantic feature information of the first text sample stored in the system, and selecting a first sample text with higher similarity between the semantic feature information and the semantic feature information of the target text (or the target word segmentation); and finally recommending the related information according to the selected first text sample. The semantic determining model is mainly used for extracting semantic features of a target text and a target word, is an artificial intelligence-based machine learning model, can be obtained through training by a certain training method, and is used for presetting operation logic of the model in an information recommendation system.

For training of the semantic determination model, training is required according to a certain training sample, wherein the training sample comprises the following steps: the sample text and the corresponding semantic annotation information of the sample text, so that a large amount of corpus with exact semantics (i.e. the sample text with the semantic annotation information) is needed to train a more accurate semantic determination model.

The artificial intelligence (Artificial Intelligence, AI) is a theory, method, technique and application system that simulates, extends and expands human intelligence, senses environment, obtains knowledge and uses knowledge to obtain optimal results using a digital computer or a machine controlled by a digital computer. In other words, artificial intelligence is an integrated technology of computer science that attempts to understand the essence of intelligence and to produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence, i.e. research on design principles and implementation methods of various intelligent machines, enables the machines to have functions of sensing, reasoning and decision.

The artificial intelligence technology is a comprehensive subject, and relates to the technology with wide fields, namely the technology with a hardware level and the technology with a software level. Artificial intelligence infrastructure technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and other directions.

It can be seen that, in the method of this embodiment, the information recommendation system matches the association model by using the target text and the first target association text and the second target association text corresponding to the target word respectively, so as to obtain a first sample text and a first association text that are matched, and further, perform information recommendation based on the first sample text and the first association text. In the process, the text does not need to be subjected to semantic analysis, but the associated text related to the text in terms of semantics, application scenes and the like can be considered, so that the limitation caused in the semantic analysis process can be avoided, more comprehensive text related to the target text, namely the matched first sample and the first associated text, can be provided, and the comprehensive and accurate recommendation of the target text is realized.

Another embodiment of the present invention provides an information recommendation method, mainly implemented by an information recommendation system, where the method in this embodiment is different from the method shown in fig. 2 in that the method in this embodiment mainly performs information recommendation according to a text map, and in the method shown in fig. 2, recommendation of target text related information according to an association model, a flowchart of the method in this embodiment is shown in fig. 3, and includes:

Step 201, obtaining a target text, and segmenting the target text to obtain target segmented words of the target text.

The user can input a text through an interface provided by the information recommendation system, and after the information recommendation system acquires the text input by the user, the text is used as a target text, and the information recommendation method flow of the embodiment is initiated. Or, the user may input a section of voice through an interface provided by the information recommendation system, and after the information recommendation system obtains the section of voice input by the user, the section of voice is converted into text, and the converted text is used as a target text, so as to initiate the information recommendation method flow of the embodiment.

Step 202, acquiring a text map, wherein the text map comprises a plurality of second sample texts and information whether association is carried out between any two second sample texts based on the topic type.

The information recommendation system may obtain the text map before initiating the information recommendation method flow of the embodiment, and store the text map in the local storage, and after initiating the information recommendation method flow of the embodiment, the information recommendation system may directly extract the text map from the local storage.

In a specific application example, when the information recommendation system acquires a text map, a historical search record obtained by the sample recommendation system can be analyzed to obtain the text map, specifically, the information recommendation system can acquire search operation information in the sample recommendation system, and the search operation information comprises a search text; acquiring sample word segmentation corresponding to the search text, and acquiring characteristic information corresponding to the search text and the sample word segmentation respectively, wherein the characteristic information can comprise ranking information, heat information, associative text and the like; performing topic clustering on the search text and the sample word according to the characteristic information respectively corresponding to the search text and the sample word to obtain a second sample text with multiple topic types; and correlating any two second sample texts in the same theme type to obtain a text map.

When the topic clustering is performed on the search text and the sample word according to the feature information respectively corresponding to the search text and the sample word, the feature information respectively corresponding to the search text and the sample word can be subjected to feature quantization, and then the topic clustering algorithm is adopted to perform the topic clustering. The topic clustering algorithm may include topic clustering algorithms such as latent semantic indexing (Latent Semantic Indexing, LSI), also called latent semantic analysis (Latent Semantic Analysis, LSA), etc., and may find the latent semantics of the second sample text, and cluster the second sample text belonging to the same semantics, for example, "car" and "automatic" have the same meaning, and may perform topic clustering. Wherein the subject of the second sample text refers to the attribute of the second sample text, and the subject of the "car" is a vehicle or the like.

Further, the information recommendation system determines similarity parameters between the second sample text of any two of the plurality of topic types; and associating the second sample texts of any two theme types, wherein the similarity parameters meet preset conditions (such as the similarity parameters are larger than preset values). The similarity parameter may refer to a parameter such as similarity for describing similarity between two texts, so that a second sample text in two topic types with higher similarity may be associated with each other.

And 203, matching the target text and the target word thereof with a second sample text in the text map respectively to obtain a second sample text associated with the target text and the target word.

Specifically, if the target text (or target word) matches a certain second sample text in the text map, the second sample text associated with the target text and the target word specifically includes: some second sample text that matches the target text or target word, and further includes: other second sample text associated with a certain second sample text in the text map based on the subject.

For example, fig. 4 shows a text map obtained by the information recommendation system, where the text map includes a plurality of nodes, each node corresponds to a second sample text, and edges between the nodes represent topic-based associations between the corresponding second sample texts. For example, the second sample text represented by the nodes A, B, C and D belongs to the topic type 1, the second sample text represented by the nodes E and F belongs to the topic type 2, the second sample text represented by the nodes G, H and I belongs to the topic type 3, further, the similarity parameters between the topic type 1 and the second sample text represented by the topic type 2 satisfy the preset condition, the two topic types are associated, and the topic types 1 and 2 are not associated with the topic type 3 respectively. Thus, when the target text entered by the user or its corresponding target word match the second sample text represented by node E in topic type 2, then the associated second sample text that is ultimately obtained includes: the second sample text represented by nodes A, B, C, D, E and F.

And 204, recommending related subjects to the target text according to the second sample text obtained by matching in the step 203.

Specifically, the information recommendation system may directly output the second sample text obtained by matching in the step 203, so as to display the second sample text to the user; alternatively, the information recommendation system may apply the second sample text obtained by the matching to a recommendation process of some specific information, such as program recommendation, news or video information recommendation.

In this way, the target text and the target word thereof are respectively matched with the text map, so that a second sample text matched with the target text or the target word thereof based on the topic type can be obtained, and the direct semantics and the potential semantics of the target text and the target word can be found, thereby realizing more comprehensive topic recommendation.

In order to further improve accuracy of the recommended information, the information recommendation system can obtain semantic feature information corresponding to the search text and the sample word thereof respectively according to the search text, the sample word and a preset semantic determination model in the sample recommendation system, namely semantic feature information of the second sample text, then clustering is carried out according to similarity among the semantic feature information of the second sample text, and the obtained text map is adjusted based on a clustering result. In addition, when a user inputs a target text in the information recommendation system in real time, the information recommendation system can acquire semantic feature information corresponding to the target text and the target word according to the target text, the target word and a preset semantic determination model; and then, selecting a second sample text with higher similarity between the semantic feature information and the semantic feature information corresponding to the target text and the target segmentation respectively from the text map. The semantic determining model is mainly used for extracting semantic features of a target text and a target word, is an artificial intelligence-based machine learning model, can be obtained through training by a certain training method, and is used for presetting operation logic of the model in an information recommendation system.

The information recommending method of the present invention is described below with a specific application example, and the method in this embodiment is mainly applied to recommending related applications according to a target text input by a user, as shown in fig. 5, and the method in this embodiment mainly includes the following two parts:

(1) And acquiring an association model, and presetting the association model into the information recommendation system.

In step 301, the information recommendation system acquires search operation information in the sample recommendation system, where the search operation information includes a search text and a search result of the sample recommendation system on the search text. Wherein the sample recommendation system may be a certain application store.

In step 302, the information recommendation system performs word segmentation on the search text to obtain sample word segmentation, and eliminates nonsensical sample word segmentation in the sample word segmentation.

In step 303, the information recommendation system obtains feature information corresponding to the search text and the sample word, which may specifically include heat information, ranking information, and association text, so as to obtain an association model, where feature information of a first sample text included in the association model is the feature information corresponding to the search text and the sample word, and first sample text included in the association model is the search text and the sample word thereof.

When the information recommendation system obtains the associated text corresponding to the search text and the sample word, the information recommendation system mainly obtains the associated text according to the search result of the sample recommendation system on the search text, and the obtaining method is described in the above embodiment and is not repeated herein.

In step 304, the information recommendation system extracts semantic features of the search text and the sample word, for example, extracts semantic features by using a semantic determination model such as word2vec, and stores extracted semantic feature information, which includes semantic feature information corresponding to the search text and the sample word (i.e., the first sample text). Thus, the feature information of each first text sample in the association model also includes semantic feature information corresponding to each first text sample.

(2) And recommending information to the target text input by the user in real time and the acquired association model.

In step 305, the information recommendation system provides a user input interface such that the user can input the target text in the user input interface, and the user can also select the language type of the information recommended by the information recommendation system in the user input interface.

Step 306, the information recommendation system performs word segmentation on the target text to obtain target word segmentation, and obtains a first target association text, a second target association text and the like which respectively correspond to the target text and the target word segmentation.

Specifically, when the information recommendation system obtains the first and second target associated texts, the information recommendation system can be realized by adopting an associator, and vocabulary characters are increased and decreased and supplemented for the target text and the target word, for example, the associated texts of 'more than one spelling' are 'more than one spelling', 'more than one spelling' and the like.

In step 307, the information recommendation system matches the target text, the target word segment, the first target associated text, and the second target associated text with the obtained association model to obtain a first associated text and a first sample text that are matched with the target text, the target word segment, the first target associated text, and the second target associated text. In fig. 5, only the matching of the target associated text with the associated model is shown, and the matching of the target text and the target word with the associated model is not shown.

In step 308, the information recommendation system may further determine semantic feature information corresponding to the target text and the target word according to the target text, the target word and a preset semantic determination model, and then match the determined semantic feature information with the semantic feature information of the first sample obtained in step 304, to obtain a first sample text with higher similarity between the semantic feature information corresponding to the target text and the target word.

In step 309, the information recommendation system performs information recommendation on the target text according to the first associated text and the first sample text obtained in step 307 and the first sample text obtained in step 308, for example, directly outputting the first sample text and the first associated text for displaying to the user.

For example, if the user wants to find japanese text, but the user is familiar with english, the user may select a language type to be queried through a user input interface provided by the information recommendation system, and input a target text "brawl star", the information recommendation system may use the method of the embodiment to recommend related information to the target text. As shown in fig. 6, the information recommendation system may display a large number of japanese synonyms corresponding to the target text "brawl star", such as "low-level" and "ぶ allowance" and so on, which brings great convenience to users who do not understand small languages.

The information recommending method of the present invention is described below with another specific application example, and the method in this embodiment is mainly applied to recommending related applications according to a target text input by a user, as shown in fig. 7, and then the method in this embodiment mainly includes the following two parts:

(1) And acquiring a text map, and presetting the text map into the information recommendation system.

In step 401, the information recommendation system acquires search operation information in the sample recommendation system, where the search operation information includes a search text. Wherein the sample recommendation system may be a certain application store.

In step 402, the information recommendation system performs word segmentation on the search text to obtain sample word segmentation, and eliminates nonsensical sample word segmentation in the sample word segmentation.

In step 403, the information recommendation system obtains feature information corresponding to the search text and the sample word, which may include heat information, ranking information, and associative text.

And step 404, the information recommendation system performs topic clustering on the search text and the sample word according to the characteristic information respectively corresponding to the user search and the sample word to obtain a second sample text with multiple topic types.

In one case, the information recommendation system can adopt an LSI algorithm to perform topic clustering, so that implicit topics of the search text and sample word thereof can be found, specifically, the information recommendation system can perform feature quantization on feature information corresponding to the search text and the sample word respectively to obtain quantized features; then mapping the quantized features to a new semantic space by using a singular value decomposition (Singular Value Decomposition, SVD) dimension reduction method to obtain semantic space features; and finally, carrying out semantic clustering based on semantic space features respectively corresponding to the search text and the sample word segmentation.

For example, for a matrix AA of m×nm×n, one can decompose into three matrices shown in the following equation 1:

Am×n＝Um×mΣm×nVTn×n (1)

sometimes to reduce the dimension of the matrix to k, the decomposition of SVD can be approximated as follows in equation 2:

Am×n≈Um×kΣk×kVTk×n(2)

wherein, assuming that m search texts are provided, each search text has n sample word segments, aij represents characteristic information of the j sample word segments of the i search text, and the characteristic information is most commonly a Term Frequency-inverse document Frequency (TF-IDF) value; k is the number of topic types, generally less than the number of search text; after SVD decomposition, uil represents the relevance of the ith search text and the ith topic type, vjm represents the relevance of the jth sample word and the mth sample word, and Σlm represents the relevance of the ith topic type and the mth sample word.

The SVD dimension reduction method can be regarded as finding irrelevant index variables (factors) from quantized features, mapping the quantized features to semantic space, and the obtained feature information contains the search text and the implicit semantics of sample word segmentation thereof, so that the SVD dimension reduction method is accurate. If any two text with dissimilar quantized features are likely to be similar in semantic space, the method of the embodiment can find the hidden semantics of the text, so that the feature information of the obtained semantic space contains the hidden semantics of the text.

In the case of performing the topic clustering, a clustering method other than the LSI algorithm may be used, and the topic clustering may not be performed.

Further, in order to obtain the text map more accurately, in the process of executing the step 404, the information recommendation system may perform a "pruning" operation on some data in the semantic space after mapping the quantized features of the search text and the sample word to the semantic space, specifically, if the number of elements with zero value in the semantic space corresponding to a certain search text or sample word is greater than a preset value, the semantic space feature value of the search text or sample word is removed.

In addition, after obtaining the second sample texts of the multiple topic types, the information recommendation system can perform a topic removal operation, namely removing some second sample texts in some topic types. In particular, second sample text included in any topic type that is inconsistent with that topic type may be removed.

Step 405, the information recommendation system correlates any two second sample texts in the same topic type, and determines similar parameters between the second sample texts in any two topic types in multiple topic types; and correlating the second sample texts of any two theme types with similar parameters meeting preset conditions (such as the similar parameters are larger than preset values), further obtaining a text map, and storing the text map in an information recommendation system.

In step 406, the information recommendation system further performs semantic feature extraction on the search text and the sample word, for example, performs semantic feature extraction by using a semantic determination model such as word2vec, clusters the search text and the sample word based on similarity between semantic feature information, and adjusts the text map obtained in step 405 according to the clustering result.

(2) And recommending information to the target text input by the user in real time and the acquired text map.

In step 407, the information recommendation system may provide a user input interface, such that the user may input the target text in the user input interface, and the user may also select a language type of the information recommended by the information recommendation system in the user input interface.

In step 408, the information recommendation system performs word segmentation on the target text to obtain a target word.

And 409, the information recommendation system respectively matches the target text and the target word with the obtained text map to obtain a second sample text matched with the target text or the target word.

In step 410, the information recommendation system recommends the program of the related subject to the target text according to the second sample text acquired in step 409.

For example, if a user wants to find japanese text, but the user is familiar with english, the user may select a language type to be queried through a user input interface provided by the information recommendation system, and input a target text "PUBG", and the information recommendation system may use the method of the embodiment to perform topic recommendation on the target text. As shown in fig. 8a, the information recommendation system may display other application programs belonging to the same theme type as the target text "PUBG", such as "wild action", "brawl star" and "moba", and may also display applications of related theme such as FPS, and further include many japanese application programs, so as to bring great convenience for users who do not understand small languages.

In addition, the method of the embodiment has a certain advantage in terms of semantic debug, for example, if the target text input by the user is wrong, but if the whole text is not affected, reasonable related information can be recommended, for example, as shown in fig. 8b, when the user inputs "start" when inputting the target text, the target text input by the user is "brawl start", and the information recommendation system can reasonably display information actually related to "brawl start". And if the current information recommendation system is taken as a sample recommendation system, a text map as shown in fig. 8c can be obtained, wherein the text associated with the topic of "brawl start" is actually the text associated with the topic of "brawl start".

Therefore, through the method of the embodiment, semantic understanding can be reasonably carried out on each text, the effect is good in information recommendation, and especially when small-language information recommendation is carried out, the restriction of a semantic analysis process is avoided as semantic analysis is not needed, and information recommendation can be accurately carried out.

The information recommending method of the present invention is described in another specific application example below, and the method in this embodiment is mainly applied to inputting english phrases by a user into an information recommending system, and the information recommending system uses the english phrases as target texts to recommend information, as shown in fig. 9, where the method in this embodiment mainly includes the following two parts:

in step 501, the information recommendation system provides a user input interface, so that a user can input an english phrase, such as "brawl start", in the user input interface, and the user can also select a language type of the information recommended by the information recommendation system, specifically, japanese in the user input interface.

In step 502, the information recommendation system uses the english phrase input by the user as a target text, and performs word segmentation on the target text to obtain target word segments, namely, "brawl" and "start", and may also obtain a first target associated text and a second target associated text corresponding to the target text and the target word segment, respectively, where the second target associated text, such as "start", may include "star" and the like.

Further, a text map and an association model may be preset in the information recommendation system in advance, if the information recommendation system determines the association model to be matched, and the following step 503 is performed; if the information recommendation system determines a text atlas to be matched, then the following step 504 is performed

In step 503, the information recommendation system matches "brawl start", "brawl", "start", the first target associated text and the second target associated text with the association model, respectively, to obtain a first sample text and a first associated text that are matched with "brawl start", "brawl", "start", the first target associated text and the second target associated text, where the first sample text and the first associated text include related japanese text, and the first sample text and the first associated text that are matched can be directly output. Such as the information shown in fig. 8b above.

In step 504, the information recommendation system matches "brawl start", "brawl" and "start" with the text map, respectively, to obtain second sample text matched with "brawl start", "brawl" and "start", where related japanese text is included, and the matched second sample text can be directly output. Such as the information shown in fig. 8b above.

The embodiment of the invention also provides an information recommendation system, the structure schematic diagram of which is shown in fig. 10, and the system specifically may include:

a text acquisition unit 10 for acquiring a target text.

The text obtaining unit 10 is further configured to segment the target text to obtain a target segment corresponding to the target text, and obtain a first target associated text corresponding to the target text and a second target associated text corresponding to the target segment.

The model obtaining unit 11 is configured to obtain an association model, where a plurality of first sample texts and a plurality of feature information corresponding to the first sample texts are stored, and the feature information includes a first association text.

The model obtaining unit 11 is specifically configured to obtain search operation information in a sample recommendation system, where the search operation information includes a search text and a search result of the sample recommendation system on the search text; acquiring sample word segmentation corresponding to the search text; according to the search results corresponding to the search text, counting the associated texts corresponding to the sample word segmentation and the search text respectively, wherein the associated texts corresponding to the sample word segmentation and the search text respectively are the first associated text in the associated model, and the sample word segmentation and the search text are the first sample text in the associated model.

Wherein, the search result comprises: the model obtaining unit 11 calculates, according to a search result corresponding to the search text, association texts corresponding to the sample word and the search text, when the ranking information and the heat information of any one piece of first search information satisfy a preset condition, and when any one piece of first search information includes the association text corresponding to the search text, determines the association text corresponding to the search text according to any one piece of first search information; when the ranking information and the heat information of any piece of second search information meet preset conditions, and the any piece of second search information contains the associated text corresponding to the sample word, determining the associated text corresponding to the sample word according to any piece of second search information.

The model obtaining unit 11 is further configured to determine heat information and ranking information corresponding to the sample word segmentation and the search text, where the feature information of the first sample text further includes heat information and ranking information corresponding to the sample word segmentation and the search text, respectively.

And a matching unit 12, configured to match the target text, the target word, the first target associated text, and the second target associated text acquired by the text acquiring unit 10 with the first sample text and the first associated text in the association model acquired by the model acquiring unit 11, respectively, to obtain the first sample text and the first associated text that are matched with the target text, the target word, the first target associated text, and the second target associated text.

And a recommending unit 13, configured to recommend information to the target text according to the first sample text and the first associated text obtained by matching by the matching unit 12.

The recommending unit 13 is specifically configured to match the target text, the target word, the first target associated text, and the second target associated text with each first sample text and each first associated text in the association model, so as to obtain a matched first sample text and a first associated text; and selecting the first sample text and the first association text of which the ranking information and the heat information meet preset conditions from the matched first sample text and the first association text.

Further, the information recommendation system in this embodiment further includes:

A map acquisition unit 14, configured to acquire a text map, where the text map includes a plurality of second sample texts and information about whether or not association is performed between any two second sample texts based on a topic type; the recommending unit 13 is further configured to match the target text and the target word thereof with the text map acquired by the map acquiring unit 14, respectively, to obtain a second sample text associated with the target text and the target word; the map obtaining unit 14 is further configured to recommend a related subject to the target text according to the second sample text obtained by matching by the recommending unit 13.

Wherein, the map acquisition unit 14 is specifically configured to acquire search operation information in the sample recommendation system, where the search operation information includes a search text; acquiring sample word segmentation corresponding to the search text and acquiring characteristic information corresponding to the search text and the sample word segmentation respectively; performing topic clustering on the search text and the sample word according to the characteristic information respectively corresponding to the search text and the sample word to obtain a second sample text with multiple topic types; and associating any two second sample texts in the same theme type.

The map acquisition unit 15 is further configured to determine a similarity parameter between the second sample text of any two topic types in the plurality of topic types; and correlating the second sample texts of any two theme types with similar parameters meeting preset conditions.

As can be seen, in the information recommendation system of the present embodiment, the matching unit 12 matches the association model by using the first target association text and the second target association text corresponding to the target text and the target word thereof, so as to obtain the matched first sample text and first association text, and the recommendation unit 13 recommends information based thereon. In the process, the text does not need to be subjected to semantic analysis, but the associated text related to the text in terms of semantics, application scenes and the like can be considered, so that the limitation caused in the semantic analysis process can be avoided, more comprehensive text related to the target text, namely the matched first sample and the first associated text, can be provided, and the comprehensive and accurate recommendation of the target text is realized.

The embodiment of the present invention further provides a terminal device, whose structure schematic diagram is shown in fig. 11, where the terminal device may generate relatively large differences due to different configurations or performances, and may include one or more central processing units (central processing units, CPU) 20 (e.g., one or more processors) and a memory 21, and one or more storage media 22 (e.g., one or more mass storage devices) storing application programs 221 or data 222. Wherein the memory 21 and the storage medium 22 may be transitory or persistent. The program stored in the storage medium 22 may include one or more modules (not shown), each of which may include a series of instruction operations in the terminal device. Still further, the central processor 20 may be arranged to communicate with the storage medium 22 and execute a series of instruction operations in the storage medium 22 on the terminal device.

Specifically, the application program 221 stored in the storage medium 22 includes an application program for information recommendation, and the program may include the text acquisition unit 10, the model acquisition unit 11, the matching unit 12, the recommendation unit 13, and the map acquisition unit 14 in the above-described information recommendation system, which will not be described in detail herein. Still further, the central processor 20 may be configured to communicate with the storage medium 22 and execute a series of operations corresponding to the application program recommended by the information stored in the storage medium 22 on the terminal device.

The terminal device may also include one or more power supplies 23, one or more wired or wireless network interfaces 24, one or more input/output interfaces 25, and/or one or more operating systems 223, such as Windows ServerTM, mac OS XTM, unixTM, linuxTM, freeBSDTM, or the like.

The steps performed by the information recommendation system described in the above-described method embodiment may be based on the structure of the terminal device shown in fig. 11.

Another aspect of the embodiments of the present invention also provides a computer-readable storage medium storing a plurality of computer programs adapted to be loaded by a processor and to perform a data transfer method as performed by the information recommendation system described above.

the memory is used for storing a plurality of computer programs, and the computer programs are used for being loaded by the processor and executing the data transfer method executed by the information recommendation system; the processor is configured to implement each of the plurality of computer programs.

Those of ordinary skill in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be implemented by a program to instruct related hardware, the program may be stored in a computer readable storage medium, and the storage medium may include: read Only Memory (ROM), random Access Memory (RAM), magnetic or optical disks, and the like.

The foregoing describes in detail a method, a system, a storage medium and a terminal device for recommending information provided by the embodiments of the present invention, and specific examples are applied to describe the principles and implementations of the present invention, where the descriptions of the foregoing embodiments are only used to help understand the method and core ideas of the present invention; meanwhile, as those skilled in the art will have variations in the specific embodiments and application scope in accordance with the ideas of the present invention, the present description should not be construed as limiting the present invention in view of the above.

Claims

1. An information recommendation method, comprising:

acquiring a target text;

word segmentation is carried out on the target text so as to obtain target word segmentation corresponding to the target text;

acquiring an association model, wherein a plurality of first sample texts and characteristic information respectively corresponding to the plurality of first sample texts are stored in the association model, and the characteristic information comprises first association texts corresponding to the first sample texts;

matching the target text, the target word segmentation, the first target association text and the second target association text with a first sample text and a first association text in the association model respectively to obtain the first sample text and the first association text matched with the target text, the target word segmentation, the first target association text and the second target association text;

according to the first sample text and the first association text obtained by matching, recommending information to the target text;

the acquiring the association model specifically comprises the following steps:

acquiring search operation information in a sample recommendation system, wherein the search operation information comprises a search text and a search result of the sample recommendation system on the search text;

Acquiring sample word segmentation corresponding to the search text;

according to the search results corresponding to the search text, counting the associated texts corresponding to the search text and the sample word segmentation search text respectively, wherein the sample word segmentation and the search text are the first sample text stored in the associated model, and the associated text corresponding to the sample word segmentation and the search text respectively is the first associated text stored in the associated model.

2. The method of claim 1, wherein the search results comprise: the statistics of the associated text corresponding to the search text and the sample word respectively according to the search result corresponding to the search text specifically includes:

when ranking information and heat information of any piece of first search information meet preset conditions, and the any piece of first search information contains associated text corresponding to the search text, determining the associated text corresponding to the search text according to any piece of first search information;

if the ranking information and the heat information of any piece of second search information meet preset conditions, and the any piece of second search information contains the associated text corresponding to the sample word, determining the associated text corresponding to the sample word according to any piece of second search information.

3. The method of claim 1, wherein the characteristic information further includes heat information and ranking information corresponding to a plurality of the first samples, respectively, the obtaining the association model further comprising:

and determining the heat information and ranking information respectively corresponding to the sample word segmentation and the search text, wherein the characteristic information of the first sample text also comprises the heat information and ranking information respectively corresponding to the search text.

4. The method of claim 3, wherein the matching the target text, the obtained target word, the first target associated text, and the second target associated text with the first sample text and the first associated text in the association model respectively, to obtain the first sample text and the first associated text that match the target text, the target word, and the first target associated text and the second target associated text, specifically comprises:

matching the target text, the target word segmentation and the target association text with each first sample text and each first association text in the association model respectively to obtain a matched first sample text and a first association text;

and selecting the first sample text and the first association text of which the ranking information and the heat information meet preset conditions from the matched first sample text and the first association text.

5. The method of any one of claims 1 to 4, further comprising:

acquiring a text map, wherein the text map comprises a plurality of second sample texts and information whether the association between any two second sample texts is performed based on the topic type;

matching the target text and the target word thereof with a second sample text in the text map respectively to obtain a second sample text associated with the target text and the target word;

and recommending related subjects to the target text according to the second sample text obtained by matching.

6. The method of claim 5, wherein the obtaining a text map specifically comprises:

acquiring search operation information in a sample recommendation system, wherein the search operation information comprises a search text;

acquiring sample word segmentation corresponding to the search text and acquiring characteristic information corresponding to the search text and the sample word segmentation respectively;

performing topic clustering on the search text and the sample word according to the characteristic information respectively corresponding to the search text and the sample word to obtain a second sample text with multiple topic types;

and correlating any two second sample texts in the same theme type to obtain the text map.

7. The method of claim 6, wherein after obtaining the second sample text for the plurality of topic types, further comprising:

determining a similarity parameter between second sample texts of any two topic types in the topic types;

and correlating the second sample texts of any two theme types with similar parameters meeting preset conditions.

8. An information recommendation system, comprising:

the text acquisition unit is used for acquiring a target text;

the model acquisition unit is used for acquiring an association model, wherein the association model comprises a plurality of first sample texts and characteristic information corresponding to the plurality of first sample texts, and the characteristic information comprises first association texts; the model acquisition unit is specifically used for acquiring search operation information in a sample recommendation system, wherein the search operation information comprises a search text and a search result of the sample recommendation system on the search text; acquiring sample word segmentation corresponding to the search text; counting the associated texts corresponding to the search text and the sample word segmentation search text respectively according to the search results corresponding to the search text, wherein the sample word segmentation and the search text are the first sample text stored in the associated model, and the associated text corresponding to the sample word segmentation and the search text respectively is the first associated text stored in the associated model;

9. The system as recited in claim 8, further comprising:

the map acquisition unit is used for acquiring a text map, wherein the text map comprises a plurality of second sample texts and whether correlation information is carried out between any two second sample texts based on the topic type;

the matching unit is further used for respectively matching the target text and the target word thereof with a second sample text in the text map to obtain a second sample text associated with the target text and the target word;

and the recommending unit is also used for recommending the related subject to the target text according to the second sample text obtained by matching.

10. A computer readable storage medium, characterized in that the computer readable storage medium stores a plurality of computer programs adapted to be loaded by a processor and to perform the information recommendation method according to any one of claims 1 to 7.

11. A terminal device comprising a processor and a memory;

the memory is used for storing a plurality of computer programs for loading and executing the information recommendation method according to any one of claims 1 to 7 by a processor; the processor is configured to implement each of the plurality of computer programs.