CN116049379A - Knowledge recommendation method, knowledge recommendation device, electronic equipment and storage medium - Google Patents

Abstract

The disclosure provides a knowledge recommendation method, relates to the technical field of artificial intelligence, and particularly relates to the technical fields of natural language processing, deep learning, knowledge graph and intelligent recommendation. The specific implementation scheme is as follows: determining knowledge demand information of the target object according to the text material in the target object working scene; generating a target object image according to the knowledge demand information; determining a target knowledge base corresponding to the role type of the target object from a plurality of professional knowledge bases; recalling a knowledge resource list from the target knowledge base according to the knowledge demand information; sequencing the knowledge resources in the knowledge resource list according to the target object portrait to obtain a knowledge recommendation list; and outputting a knowledge recommendation list. The disclosure also provides a knowledge recommendation device, an electronic device and a storage medium.

Description

Knowledge recommendation method, knowledge recommendation device, electronic equipment and storage medium

Technical Field

The present disclosure relates to the field of artificial intelligence, and in particular, to the fields of natural language processing, deep learning, knowledge maps, and intelligent recommendation. More particularly, the disclosure provides a knowledge recommendation method, a knowledge recommendation device, an electronic device and a storage medium.

Background

Knowledge management refers to the process of acquiring, storing, learning, sharing and innovating knowledge by an organization as a whole, and aims to improve the productivity of knowledge workers in the organization and the core competitiveness of the organization.

The knowledge recommendation system can realize standardization and unitization of knowledge management, establish sufficient links of people and knowledge, people and people, and enable knowledge to be multiplexed and flowed maximally in work, and promote organization innovation and efficiency improvement. The knowledge recommendation system is therefore the core of knowledge management.

Disclosure of Invention

The disclosure provides a knowledge recommendation method, a knowledge recommendation device, knowledge recommendation equipment and a storage medium.

According to a first aspect, there is provided a knowledge recommendation method, the method comprising: determining knowledge demand information of the target object according to the text material in the target object working scene; generating a target object image according to the knowledge demand information; determining a target knowledge base corresponding to the role type of the target object from a plurality of professional knowledge bases; recalling a knowledge resource list from the target knowledge base according to the knowledge demand information; sequencing the knowledge resources in the knowledge resource list according to the target object portrait to obtain a knowledge recommendation list; and outputting a knowledge recommendation list.

According to a second aspect, there is provided a knowledge recommendation device, the device comprising: the first determining module is used for determining knowledge demand information of the target object according to the text materials in the target object working scene; the first generation module is used for generating a target object portrait according to knowledge demand information; the second determining module is used for determining a target knowledge base corresponding to the role type of the target object from a plurality of professional knowledge bases; the recall module is used for recalling a knowledge resource list from the target knowledge base according to the knowledge demand information; the ordering module is used for ordering the knowledge resources in the knowledge resource list according to the target object portrait to obtain a knowledge recommendation list; and the output module is used for outputting the knowledge recommendation list.

According to a third aspect, there is provided an electronic device comprising: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform a method provided in accordance with the present disclosure.

According to a fourth aspect, there is provided a non-transitory computer readable storage medium storing computer instructions for causing a computer to perform a method provided according to the present disclosure.

According to a fifth aspect, there is provided a computer program product comprising a computer program stored on at least one of a readable storage medium and an electronic device, which, when executed by a processor, implements a method provided according to the present disclosure.

It should be understood that the description in this section is not intended to identify key or critical features of the embodiments of the disclosure, nor is it intended to be used to limit the scope of the disclosure. Other features of the present disclosure will become apparent from the following specification.

Drawings

The drawings are for a better understanding of the present solution and are not to be construed as limiting the present disclosure. Wherein:

FIG. 1 is a schematic diagram of an exemplary system architecture to which knowledge recommendation methods and apparatus may be applied, according to one embodiment of the present disclosure;

FIG. 2 is a flow chart of a knowledge recommendation method, according to one embodiment of the present disclosure;

FIG. 3 is a schematic diagram of determining knowledge demand information, according to one embodiment of the present disclosure;

FIG. 4 is a schematic diagram of determining knowledge demand information, according to one embodiment of the disclosure;

FIG. 5 is a schematic diagram of a knowledge network and a method of constructing a specialized knowledge base, according to an embodiment of the disclosure;

FIG. 6 is a schematic diagram of a keyword index structure that generates a expertise store according to an embodiment of the disclosure;

FIG. 7 is an overall framework diagram of a knowledge recommendation method, according to one embodiment of the disclosure;

FIG. 8 is a block diagram of a knowledge recommendation device, according to one embodiment of the disclosure;

fig. 9 is a block diagram of an electronic device of a knowledge recommendation method, according to one embodiment of the disclosure.

Detailed Description

Exemplary embodiments of the present disclosure are described below in conjunction with the accompanying drawings, which include various details of the embodiments of the present disclosure to facilitate understanding, and should be considered as merely exemplary. Accordingly, one of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

Traditional knowledge recommendation methods include recommendation algorithms based on content similarity, or collaborative filtering algorithms based on user behavior. The recommendation algorithm based on content similarity is to recommend similar articles to the user according to articles clicked/browsed/collected by the user. Collaborative filtering algorithm based on user behavior is to analyze user behavior, determine which are similar users, and recommend articles clicked/browsed/collected by the similar users to the users.

Recommendation algorithms based on content similarity or collaborative filtering algorithms based on user behaviors tend to be implicit representations of the points of interest of the user, even if the points of interest of the user are not calculated, knowledge resources such as articles, videos and the like which are possibly interested are recommended to the user only according to the similar user behaviors, and therefore recommendation effects are poor.

A knowledge recommendation method calculates high-frequency keywords related to user behaviors through word frequency statistics and other methods, and uses the high-frequency keywords as attention points of users, so that knowledge resources interested by the users are recalled according to the attention points.

The attention points are calculated through a word frequency statistical method, the scheme is simple, the operability is strong, but only attention points with high frequency can be obtained. Fewer points of interest are found for long tails and maintaining a list of points of interest is costly. In addition, only the attention points of the closed set can be obtained, the recording timeliness of the new attention points is poor, and the requirement of a user for acquiring new knowledge cannot be met.

In the technical scheme of the disclosure, the related processes of collecting, storing, using, processing, transmitting, providing, disclosing and the like of the target object personal information all accord with the regulations of related laws and regulations, and the public order is not violated.

In the technical scheme of the disclosure, the authorization or consent of the target object is acquired before the personal information of the target object is acquired or collected.

FIG. 1 is a schematic diagram of an exemplary system architecture to which knowledge recommendation methods and apparatus may be applied, according to one embodiment of the present disclosure. It should be noted that fig. 1 is only an example of a system architecture to which embodiments of the present disclosure may be applied to assist those skilled in the art in understanding the technical content of the present disclosure, but does not mean that embodiments of the present disclosure may not be used in other devices, systems, environments, or scenarios.

As shown in fig. 1, a system architecture 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 is used as a medium to provide communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired and/or wireless communication links, and the like.

The target object may interact with the server 105 via the network 104 using the terminal devices 101, 102, 103 to receive or send messages or the like. The terminal devices 101, 102, 103 may be a variety of electronic devices including, but not limited to, smartphones, tablets, laptop portable computers, and the like.

The knowledge recommendation methods provided by embodiments of the present disclosure may be generally performed by the server 105. Accordingly, the knowledge recommendation device provided by the embodiments of the present disclosure may be generally provided in the server 105. The knowledge recommendation method provided by the embodiments of the present disclosure may also be performed by a server or a server cluster that is different from the server 105 and is capable of communicating with the terminal devices 101, 102, 103 and/or the server 105. Accordingly, the knowledge recommendation apparatus provided by the embodiments of the present disclosure may also be provided in a server or a server cluster that is different from the server 105 and is capable of communicating with the terminal devices 101, 102, 103 and/or the server 105.

FIG. 2 is a flow chart of a knowledge recommendation method, according to one embodiment of the present disclosure.

As shown in fig. 2, the knowledge recommendation method 200 includes operations S210 to S260.

In operation S210, knowledge requirement information of the target object is determined according to text material in the target object work scene.

For example, the target object may be a user, and the text material of the target object in the work scene may be various forms of work material generated by the user in the work scene or the office scene. For example, work reports (e.g., weekly reports), meeting records, work related articles authored by the user, content searched by the user, and the like.

For example, entities having significance for a user's attention in a work scene, such as technology, skills, fields, topics, etc., may be mined from text material. Such as "a algorithm", "B model", "C product", etc., which are collectively referred to as points of interest. Specific problems encountered by the user in the working scene can be mined from the text material, for example, the user needs to produce an XX model investigation report, and the XX model investigation report is the specific problem of the user.

The above concerns and specific questions can be used as knowledge requirement information for the user. This knowledge requirement information is a core difficulty encountered in the user's work or an important skill to master in the work. Therefore, the knowledge demand information is extracted from the work materials of the user, so that the knowledge demand information of the user is displayed, and the knowledge resources required by the user can be recommended better.

In operation S220, a target object image is generated based on the knowledge demand information.

For example, the target object representation refers to a user representation. Basic information, knowledge demand information, affinity information, etc. of the user can be aggregated into a user portrayal and stored in a user portrayal database.

The basic information of the user may include basic information of a mailbox where the user works, a character type (e.g., technician, manager, product person, etc.) in the work, a department, etc.

Knowledge demand information may include points of interest to a user over a long or short period of time, as well as specific questions over a long or short period of time.

Affinity information may include basic information for a number of other users whose affinity to the current user is greater than a threshold (e.g., 70%). For example, the basic information of a plurality of (e.g., 50) other users that are most closely related to the current user may be recorded as affinity information of the current user centering on the current user.

In contrast to conventional knowledge recommendation systems, points of interest to users tend to be implicit, have no interpretability and lack of product controllability. The user portrayal constructed by the embodiment can show the topics, technologies, fields, specific problems and the like which are interested by the user. On one hand, the system is more interpretable, on the other hand, the system can help the user to better summarize and manage the attention points and the problems, and more systematically comb own portraits, thereby accelerating the innovation of the user

In operation S230, a target knowledge base corresponding to the character type of the target object is determined from among the plurality of specialized knowledge bases.

The professional knowledge base may be obtained by dividing a resource corpus, and the resource corpus may include news, various articles and videos disclosed on the internet, and articles and reports authored by users within the enterprise.

The role types of the user in the work may include technician, manager, product personnel, etc. To recommend appropriate knowledge resources to users of different character types, the corpus of resources can be partitioned into specialized knowledge bases (also referred to as drop-type repositories) that are suitable for use in different character types.

The plurality of specialized knowledge bases may include a research and development specialized base, a product specialized base, a management specialized base, a general base, etc., which may be a resource base containing all specialized knowledge bases. The research and development professional library can be a professional knowledge library for technicians, the product professional library can be a professional knowledge library for product personnel, and the management professional library can be a professional knowledge library for management personnel.

For example, in the case where the user's character type is a technician, the target knowledge base is a research and development professional base. Similarly, in the case where the role type of the user is manager, the target knowledge base is a management specialty base. And under the condition that the role type of the user is a product personnel, the target knowledge base is a product specialty base.

Compared with the prior art, only one resource library is constructed to record all the resources, and deep description of knowledge resources is lacked. According to the embodiment, the knowledge resources are reorganized, the professional knowledge base applicable to different role types is constructed, and high-quality resources with better pertinence can be provided for users with different role types.

In operation S240, the knowledge resource list is recalled from the target knowledge base according to the knowledge demand information.

For example, multiple recall strategies such as keyword recall and semantic recall can be adopted based on knowledge demand information, so that the quantity, relevance and richness of recalled knowledge resources are improved.

Knowledge demand information may be points of interest or specific questions, and keyword recall may be based on similarity between the points of interest or specific questions and keywords of knowledge resources in a target knowledge base. The recall of semantics may be based on the similarity between semantic vectors of points of interest or specific questions and semantic vectors of keywords of knowledge resources in the target knowledge base.

Knowledge resources recalled by the multi-way recall strategy can be fused together to serve as candidate knowledge resources, so that preparation is made for sequencing and recommending of the knowledge resources in the next step.

When the knowledge resource is recalled based on the knowledge demand information, the knowledge resource may be recalled preferentially from the professional knowledge base corresponding to the character type of the user, and if the recall amount (for example, the number of recalled articles is less than 10) is insufficient, the recall may be continued from the general library. For example, for users with technician role types, knowledge resources may be recalled from the research and development specialty library preferentially when based on knowledge demand information, and if the amount of recall is insufficient, recall may be continued from the general library.

Because the candidate scope of recommendation is limited in the level of the resource library, the difficulty of recommendation can be reduced while the suitable resources are recalled for users with different role types, and the recommended content is more controllable.

In operation S250, the knowledge resources in the knowledge resource list are ordered according to the target object representation, so as to obtain a knowledge recommendation list.

For example, the comprehensive evaluation value of the knowledge resource can be calculated based on the matching degree between the user portrait and the knowledge resource in the knowledge resource list, the matching degree between the knowledge demand information and the knowledge resource, the quality evaluation value and the timeliness evaluation value of the knowledge resource, and the like, the knowledge resource in the knowledge resource list is ordered according to the comprehensive evaluation value, and the knowledge resource in the knowledge resource list is further filtered to obtain a final knowledge recommendation list.

Knowledge resources in the knowledge resource list can be filtered based on the user portraits, for example, knowledge resources (such as chapters) authored by other users closely related to the current user are filtered according to affinity ring information in the user portraits, and because the users closely related to the current user should have similar knowledge, the knowledge resources authored by other users closely related to the current user do not need to be referred to the current user.

The order of the knowledge resources in the knowledge resource list may also be adjusted based on diversity requirements. For example, if there are adjacent similar knowledge resources in the knowledge resource list, only one of the similar knowledge resources may be reserved, and other similar knowledge resources may be removed from the knowledge resource list or placed in a later location of the knowledge resource list.

After the sorting and filtering operations, a final knowledge recommendation list can be obtained. Because the ordering and filtering of the knowledge resource list comprehensively considers the knowledge demand information, the user portrait information, the knowledge resource self information and the relation among the knowledge resource information, the knowledge recommendation list can be more accurate.

In operation S260, a knowledge recommendation list is output.

For example, the knowledge recommendation list may be sent to the user's terminal device so that the user obtains the most appropriate knowledge resources, assisting the user in enhancing the ability.

According to the embodiment of the disclosure, knowledge demand information of a user is mined according to materials in the user work scene, appropriate resources are recalled from the professional knowledge base according to the knowledge demand information, and the recommending effect and recommending efficiency of the knowledge resources are improved.

FIG. 3 is a schematic diagram of determining knowledge demand information, according to one embodiment of the disclosure.

According to an embodiment of the present disclosure, determining knowledge requirement information of a target object according to text material in a target object work scene includes: extracting a first entity word set related to knowledge from the text material; determining a plurality of entity words from the first entity word set as initial focus points according to the semantic quality and word frequency of the entity words in the first entity word set; determining at least one initial point of interest hit by content in the work report as a candidate point of interest; and ordering at least one candidate attention point according to the semantic similarity between the candidate attention point and the work report to obtain a core attention point list.

Extracting a first set of entity words related to knowledge from the text material comprises: and inputting the text material into an entity word extraction model to obtain a first entity word set.

As shown in fig. 3, text material in a user's work scene may come from work reports 301, meeting records 302, user authored articles 303, user searched content 304, and so forth. The first set of entity words 305 may be derived from a work report 301, a meeting record 302, a user authored article 303, text material generated from content 304 searched by a user, and input into an entity word extraction model 310. The entity word extraction model 310 may be trained using a collection of knowledge-related entity words. The entity words used for training include "XX Algorithm", "XX model", and so forth.

The entity-word extraction model 310 may extract a number of entity words from the work report 301, the meeting record 302, the user-authored article 303, the user-searched content 304, resulting in a first set of entity words 305. The entity words in the first set of entity words 305 have duplicate entity words and low quality entity words.

Thus, the initial set of points of interest may be further screened from the first set of entity words 305 based on semantic quality and word frequency. The semantic quality may be a degree of clarity of semantics, e.g., the entity word "XX Algorithm" is of high semantic quality compared to the "XX Algorithm question". The term frequency may be how frequently the same entity word appears in the first set of entity words 305, the higher the term frequency, the greater the probability that the entity word is a point of interest for the user. Thus, the first 50% of entity words with word frequencies greater than a threshold (e.g., 10 times) and highest semantic quality may be determined as initial points of interest, resulting in an initial set of points of interest 306.

The entity words in the initial set of points of interest 306 may contain points of interest that are commonly used by the user but are not actually related to the user, e.g., the user often uses the "XX model" in operation, but the user does not already need knowledge about the "XX model", then the "XX model" is not the core point of interest for the user. The probability that the points of interest appearing in the work report 301 are the core points of interest of the user is relatively high, so in order to extract the core points of interest of the user, the work report 301 may be used to further match the initial set of points of interest 306.

For example, the initial set of points of interest 306 may be constructed into a tree structure 307, then matched with the work report 301, and the initial set of points of interest that match successfully added to the candidate set of points of interest 308. For example, the candidate attention point set 308 is generated using, as candidate attention points, initial attention points represented by nodes hit by the content of the job report in the tree structure 307. The matching speed is high by utilizing the tree structure, and the requirement of on-line real-time calculation can be met.

For the candidate focus set 308, the candidate focus may be further ordered according to the similarity between the semantics of the candidate focus and the semantics of the work report content 301, to obtain the core focus list 309.

The present embodiment can ensure that the number of extracted entity words is sufficiently large by extracting entity words related to knowledge from the work report 301, the meeting record 302, the article 303, and the search content 304. And then, the initial focus points with high semantic quality and high word frequency are screened out according to the semantic quality and the word frequency, and then the initial focus points are further matched with the work report by utilizing the tree structure, so that the core focus points of the user can be determined, and the accuracy of determining the focus points of the user is improved.

According to an embodiment of the present disclosure, determining knowledge requirement information of a target object from text material in a target object work scene includes: and generating a plurality of knowledge requirement information corresponding to each of the plurality of character types according to the text materials in the respective working scenes of the target objects of the plurality of character types.

Because different users belong to different role types, corresponding core focus lists can be mined for users of different role types. Specifically, the mined attention points can be suitable for different roles from materials (such as materials of technicians, materials of management personnel and the like). Thus, for users of different character types, different core focus lists may be built, such as: for the product personnel, a product class core point of interest list may be constructed. An algorithm-like core point of interest list may be constructed for the technician, and so on.

According to the embodiment, the corresponding core attention point list is built aiming at the users with different role types, so that the recall effect of subsequent recall knowledge resources based on the core attention points can be further improved.

FIG. 4 is a schematic diagram of determining knowledge demand information, according to one embodiment of the disclosure.

According to an embodiment of the present disclosure, knowledge requirement information includes specific questions; according to the text material in the target object working scene, determining knowledge demand information of the target object comprises: determining a target text containing knowledge acquisition intention from the text material; and extracting the specific problem corresponding to the knowledge acquisition intention from the target text.

Determining target text from the text material that includes knowledge acquisition intent includes: and inputting the text material into an intention judging model to obtain a target text containing knowledge acquisition intention. Extracting specific questions corresponding to the knowledge acquisition intention from the target text includes: and inputting the target text into a problem generation model to generate a specific problem corresponding to the knowledge acquisition intention.

As shown in fig. 4, the text material 401 may include text in a work report, meeting record, article, search content. The intent recognition model 410 may be a deep learning model for text bi-classification. The input of the intent recognition model 410 may be text and the output is to classify the text as whether it contains a question (contains knowledge acquisition intent). The text samples of the intent recognition model 410 when trained may be positive samples containing questions and negative samples containing no questions.

The text material 401 is input into the intention recognition model 410, and the text material 401 may be classified to obtain a text 402 containing no question and a target text 403 containing a question. For example, "research XX model, wait for report of next week output," the text indicates that the user has an intention to acquire knowledge. The text may be classified by the intent recognition model 410 as target text 403 containing a question. The graph recognition model 410 can filter out large amounts of text that are not solicited, leaving only important text that may extract the user's question ready for the next question extraction.

For target text 403 containing a question, a question generation model 420 may be entered, further extracting the specific question 404. The problem generation model 420 may be a deep learning model for text generation. The question generation model 420 may perform semantic understanding on the input text, generating semantic understanding text, which may be generated from the semantics and weights of individual words or phrases in the input text. By training the question generation model 420 using text samples labeled with a particular question, the question generation model 420 may be provided with the ability to output a particular question.

For example, the target text "research XX model, the next week yield report" input problem generation model 420, the generated result "question=xx model research" may be obtained.

A closed set as opposed to a core point of interest list. The range of the specific problems mined by the embodiment is an open field, various specific problems can be generated, and the problems are more sensitive to the newly generated problems and can be recorded more timely.

It should be noted that, the core focus list and the specific problem can be used as knowledge requirement information to participate in subsequent knowledge resource recall.

The expertise library provided in this embodiment is described below with reference to fig. 5.

According to an embodiment of the present disclosure, the knowledge recommendation method further includes a step of constructing a professional knowledge base: determining knowledge labels of each of a plurality of knowledge resources; correlating the knowledge resources with the same knowledge labels to obtain a knowledge network; and dividing the knowledge network into a plurality of professional knowledge bases corresponding to the plurality of character types according to the corresponding relation between the character types and the knowledge labels.

The knowledge labels comprise at least one reading label, at least one theme label and an timeliness label; determining knowledge tags for each of a plurality of knowledge resources includes: inputting the knowledge resources into a first multi-label classification model aiming at each knowledge resource to obtain at least one reading label of the knowledge resources; inputting the knowledge resources into a second multi-label classification model to obtain at least one theme label of the knowledge resources; and determining the timeliness label of the knowledge resource according to the creation time of the knowledge resource and the behavior information aiming at the knowledge resource.

Fig. 5 is a schematic diagram of a knowledge network and a method of constructing a specialized knowledge base, according to an embodiment of the disclosure.

As shown in FIG. 5, each node in knowledge network 510 may represent a knowledge resource, and the knowledge resources may be associated with each other by a knowledge tag. For example, knowledge tags for knowledge resource 511 include "technology, technology dynamics, experience sharing", knowledge tags for knowledge resource 512 include "technology, general skills", and knowledge tags for knowledge resource 513 include "management, technology dynamics". Knowledge resource 511 and knowledge resource 512 are associated by the same label "technology", and knowledge resource 511 and knowledge resource 513 are associated dynamically by the same label "technology".

It should be noted that, to improve the overall quality of the knowledge network, the knowledge resources used to construct the knowledge network 510 may be knowledge resources that are screened out in the initial set of resources and pass the quality evaluation.

For example, building a model for assessing the overall quality of a knowledge resource may be referred to as a quality assessment model. If the knowledge resource is an article, the structure, timeliness, semantic quality of title/content and typesetting quality of the article can be used as the input of a model, and the output is the overall quality evaluation value of the article. Articles with evaluation values greater than a threshold (e.g., up to 10, threshold 5) may be used for knowledge network construction.

The embodiment can ensure that the quality of the knowledge resources recommended by the system is high, worth referencing and valuable by filtering the knowledge resources with low quality, so that the overall quality of the knowledge network 510 can be improved.

The acquisition of knowledge tags for knowledge resources is described below.

Knowledge tags may include three broad categories, reading tags, subject tags, and time-lapse tags. The reading tags are used to describe which character types of users the knowledge resource is suitable for reading, and may specifically include "technology", "product", "sales", and "management", etc. The topic labels are used to describe topic distribution of knowledge resources, and may specifically include "technical dynamics," "academic sharing," "general skills," "experience sharing," and the like. The time-lapse labels classify articles by time dimension, which may specifically include: "hot", "general", "long tail", etc.

For example, the reading tag may be extracted by a first multi-tag classification model, and the subject tag may be extracted by a second multi-tag classification model. Both the first multi-labeled classification model and the second multi-labeled classification model may be constructed based on an ERNIE (Enhanced language Representation with Informative Entities, knowledge-enhanced semantic representation model) pre-training model.

The input to the first multi-tag classification model may be article titles and content and the output may contain at least one reading tag. For example, the title and content of knowledge resource 511 may be input into a first multi-labeled text classification model, which may result in a reading label "technique".

The input to the second multi-tag classification model may be article titles and content and the output may contain at least one subject tag. For example, the title and content of knowledge resource 511 is input into a second multi-labeled text classification model, which may result in topic labels "technology dynamics", "experience sharing".

For example, the timeliness tag may be determined based on the creation time of the knowledge resource and the user behavior (praise, comment count, etc.). For example, a time-efficient tag that creates an article with a time within half a year and a praise number above a threshold (e.g., 200) is "hot". For another example, an article that is created for more than 5 years and has a praise number above a threshold (e.g., 200) has a "long tail" time-of-day tag.

In this embodiment, knowledge resources are organized and modeled based on knowledge graph technology, so that knowledge resources are associated with each other to form a knowledge network 510, which can improve recall probability, and meanwhile, multiple labels deepen understanding of content of knowledge resources, and improve recommendation correlation.

Next, in order to recommend appropriate knowledge resources to users of different character types, a plurality of expertise repositories 520 (also referred to as drop-type repositories) adapted to the different character types may be built based on the knowledge network structure 510.

The plurality of specialized knowledge bases 520 may include a research and development specialized base 521, a product specialized base 522, a management specialized base 523, a general base 524, etc., and the general base 524 may be a resource base containing all specialized knowledge bases. Research and development specialty library 521 may be a specialty knowledge base for technicians, product specialty library 522 may be a specialty knowledge base for product personnel, and management specialty library 523 may be a specialty knowledge base for management personnel.

For example, users of different persona types may be assigned corresponding knowledge tags, thereby partitioning knowledge resources in knowledge network 510 into different specialized knowledge bases.

For example, tags that may be specified for the manager include "manage", "technology dynamics", "academic share", so that articles with "manage", "technology dynamics" and "academic share" tags may be added to the management specialty library 523.

According to the method, the knowledge resources are organized and modeled based on the knowledge graph technology, the knowledge network after knowledge graph modeling records high-quality knowledge resources, and meanwhile, the knowledge resources are built into the professional knowledge base through various knowledge labels. On one hand, a high-quality vertical resource library is established for users without role types, and on the other hand, knowledge resources can be better ordered by using the knowledge graph construction capability, so that the most suitable knowledge resources are recommended for the users.

According to embodiments of the present disclosure, the first k (e.g., k is an integer greater than or equal to 1, e.g., k=3) core points of interest in the core point of interest list may be selected from a database of user portraits, or k specific questions may be selected and then appropriate knowledge resources may be matched for the user to solve the user's problem or to inspire the user's mind.

For example, multiple recall strategies such as keyword recall, semantic recall, and the like can be employed to promote the number, relevance, and richness of recalled knowledge resources based on core points of interest or issues.

Keyword recall strategies are described below.

According to an embodiment of the present disclosure, the knowledge recommendation method further includes, for each knowledge resource, extracting a second set of entity words related to the knowledge from the knowledge resource; determining at least one entity word from the second entity word set as a keyword of a knowledge resource according to the semantic quality and word frequency of the entity words in the second entity word set; and generating a keyword index structure of each specialized knowledge base according to the keywords of each knowledge resource in the specialized knowledge base aiming at each specialized knowledge base, wherein the keyword index structure comprises a plurality of keywords and a knowledge resource identification list corresponding to each keyword.

Recalling the knowledge resource list from the target knowledge base according to the knowledge demand information comprises: determining target keywords matched with knowledge demand information in a keyword index structure of a target knowledge base; and recalling the first knowledge resource list according to the knowledge resource identification list corresponding to the target keyword.

According to the embodiment of the disclosure, the keywords refer to keywords of the knowledge resource, an entity word extraction model for extracting the attention point can be multiplexed, the keywords of the knowledge resource can be extracted, a keyword index structure of the professional knowledge base is generated based on the keywords in the professional knowledge base, and recall of the knowledge resource is performed based on the keyword index structure.

Fig. 6 is a schematic diagram of a keyword index structure that generates a expertise base according to an embodiment of the disclosure.

As shown in FIG. 6, the title and content of the knowledge resource 601 are input into the entity word extraction model 610 to obtain keywords 602 of the knowledge resource 601, and there may be a plurality of keywords 602 of the knowledge resource 601. Keywords 602 may be similar to points of interest, including entities of interest such as technology, skills, domain, subject matter, etc., such as "A algorithm", "B model", "C product", etc.

For each expertise base, a large number of keywords 602 may be extracted by the entity word extraction model 610, and a keyword index structure 620, that is, an inverted index structure, of the expertise base may be constructed according to the keywords 602. The keyword index structure 620 is stored with keywords as primary keys and with the identity of the knowledge resource containing the keywords as a value.

For example, the keyword index structure 620 includes keyword 1, keyword 2, and the like, and the knowledge resource identification list of keyword 1 includes resource 1, resource 2, and the like, that is, resources 1, resource 2, and the like each include keyword 1. Similarly, the knowledge resource identification list of keyword 2 includes resource 1, resource 7, a.once-a-all, that is, resources 1, 7, etc. each contain a keyword 2.

When the key focus or the problem is used for carrying out key word recall, a target key word matched with the key focus or the problem can be determined from a key word index structure of a target knowledge base, and the knowledge resource indicated by the knowledge resource identification list corresponding to the key word is the knowledge resource to be recalled, so that a first knowledge resource list is obtained. For example, if the target keyword is the keyword 2, then the resources such as the resource 1 and the resource 7 are knowledge resources for recall.

The embodiment builds the keyword index structure, and when the key focus or the problem is used for carrying out keyword recall, the knowledge resource identification list of the key words matched with the core focus or the problem can be directly pulled back, so that the first knowledge resource list is recalled, the operation is simple, the operability and the intervention performance are strong, and the explanation is easy. And, since the entity word extraction model 610 of the present embodiment is a entity word extraction model that is multiplexed for extracting the point of interest, it can be ensured that the recalled knowledge resource has a correlation with the core point of interest.

The semantic recall strategy is described below.

Semantic recall includes Embedding recall, also called implicit recall. In order to make up for the problem that the recall amount of the keywords is too sparse, a path of implicit vector semantic recall can be added. For example, the keyword is "text classification", and when recalling is performed using the core focus "text classification", the knowledge resource list related to the "text classification" may be recalled. However, when the core focus "article classification" is used, the relevance between the "text classification" and the "article classification" cannot be identified, so that knowledge resources cannot be recalled, and the recall quantity of keywords is too sparse.

According to an embodiment of the present disclosure, recalling the knowledge resource list from the target knowledge base according to the knowledge demand information further includes: determining semantic vectors of knowledge demand information; determining semantic vectors of knowledge resources in a target knowledge base; and recalling a second knowledge resource list from the target knowledge base according to the similarity between the semantic vector of the knowledge demand information and the semantic vector of the knowledge resource.

Semantic recall refers to searching for semantic vectors of knowledge resources similar to the semantic vector of the core point of interest or problem from the target knowledge base based on the semantic vector of the core point of interest or problem, and outputting a knowledge resource list with similarity higher than a threshold (e.g., 75%) as a second knowledge resource list.

The semantic vector can be generated through a semantic vector extraction model, the semantic vector extraction model can be constructed based on an ERNIE pre-training model, the ERNIE pre-training model can accurately capture the semantics of complex sentences, and the included angle of the vectors is smaller while the semantic accuracy of the sentences is ensured.

For example, semantic vectors may be extracted for knowledge resources using a semantic vector extraction model and indexed using FAISS (Facebook AI Similarity Search, similarity vector search library). When the core attention points or the problems are used for semantic recall, the core attention points or the problems are converted into semantic vectors, then the most similar knowledge resources are retrieved from the FAISS, and a second knowledge resource list is output.

According to the embodiment of the disclosure, knowledge resources recalled by a keyword recall strategy are adopted to form a first knowledge resource list, and knowledge resources recalled by a semantic recall strategy are adopted to form a second knowledge resource list. Both the knowledge resources in the first knowledge resource list and the knowledge resources in the second knowledge resource list may be candidate knowledge resources. For candidate knowledge resources, features such as relevance, authority, diversity and the like can be comprehensively considered, and finally a knowledge recommendation list is formed and recommended to a user.

According to an embodiment of the present disclosure, sorting knowledge resources in a knowledge resource list according to a target object representation, obtaining a knowledge recommendation list includes: determining knowledge resources in the first knowledge resource list and knowledge resources in the second knowledge resource list as candidate knowledge resources; calculating a correlation evaluation value between the candidate knowledge resource and the target object portrait; calculating a matching degree evaluation value between the knowledge label of the candidate knowledge resource and the target object portrait; determining authority evaluation value, timeliness evaluation value and quality evaluation value of the knowledge resource; weighting the relevance evaluation value, the matching degree evaluation value, the authority evaluation value, the timeliness evaluation value and the quality evaluation value to obtain a comprehensive evaluation value; determining the ordering of the knowledge resources in the knowledge resource list according to the comprehensive evaluation value; and adjusting the sequencing according to the diversity requirements of the knowledge resources in the knowledge resource list to obtain a knowledge recommendation list.

In this embodiment, the relevance evaluation value, the matching degree evaluation value, the authority evaluation value, the timeliness evaluation value, and the like are comprehensively considered, and CTR (Click-Through-Rate) of each candidate knowledge resource is calculated.

The relevance evaluation values may include a relevance evaluation value between the candidate knowledge resource and the core point of interest or problem, a relevance evaluation value between the candidate knowledge resource and the user representation, and a relevance evaluation value between the user representation and the core point of interest or problem. For example, candidate knowledge resources, user portraits, core points of interest or questions may all be represented as semantic vectors, and then cosine similarities between the vectors may be calculated to obtain respective correlation evaluation values.

The matching degree evaluation value may be determined by similarity between the knowledge tag of the candidate knowledge resource and the user image. For example, the knowledge labels of the candidate knowledge resources and the user figures may be expressed as semantic vectors, and then cosine similarity between the vectors may be calculated to obtain the matching degree evaluation value.

The authority assessment value may be determined based on factors such as the level of the user authoring the candidate knowledge resource, whether team authoring is occurring, and the like. Such as a higher ranking user or team user authored article, has a higher authority assessment value.

The timeliness assessment value may be determined when determining the timeliness label of the candidate knowledge resource. For example, the timeliness evaluation values of the timeliness tags "hot", "general" and "long tail" are respectively determined while the timeliness tags are determined according to the creation time of the knowledge resource and the user behavior.

The quality assessment value may be assessed by a quality assessment model at the time of construction of the knowledge network.

Next, the relevance evaluation value, the matching degree evaluation value, the authority evaluation value, the timeliness evaluation value, and the quality evaluation value may be integrated, and the probability that the user may click on the candidate knowledge resource, that is, the CTR value may be calculated.

For example, the correlation evaluation value, the matching degree evaluation value, the authority evaluation value, the timeliness evaluation value, and the quality evaluation value may be input to LR (Logistic Regression, logistic regression model), to obtain a weight of each evaluation value. And then carrying out weighted average on the multiple evaluation values to obtain a comprehensive evaluation value. And sequencing the candidate knowledge resources according to the comprehensive evaluation value, and then adjusting sequencing according to the diversity requirement of the knowledge resources in the knowledge resource list, for example, only reserving one candidate knowledge resource in the list aiming at similar candidate knowledge resources, and finally obtaining a knowledge recommendation list.

The method and the device for implementing the knowledge recommendation list determine a final knowledge recommendation list by integrating the relevance evaluation value, the matching degree evaluation value, the authority evaluation value, the timeliness evaluation value and the quality evaluation value, and can recommend knowledge resources with high quality, high timeliness and high relevance to the user for the user so as to promote the knowledge level of the user and accelerate the innovation of enterprises.

FIG. 7 is an overall framework diagram of a knowledge recommendation method, according to one embodiment of the disclosure.

As shown in FIG. 7, the knowledge recommendation method comprises an overall framework data source 710, a knowledge demand computation module 720, a user representation database 730, a resource corpus 740, a knowledge graph construction module 750, a professional knowledge base 760, a knowledge resource recall module 770, a ranking module 780 and a knowledge recommendation list 790.

The data sources 710 include long-term data streams and real-time data streams in a user's work scenario, including in particular work reports (e.g., weekly reports, work reports, etc.), meeting records, user-authored articles, user-searched content, and the like.

The knowledge requirement calculation module 720 is configured to calculate knowledge requirement information from the data source 710. The knowledge demand computation module 720 includes a focus computation sub-module and a problem computation sub-module, wherein the focus computation sub-module is configured to extract focus points related to knowledge from the data source 710, and further screen out core focus points. The question calculation sub-module is configured to identify target text containing knowledge acquisition intent from the data source 710 using the intent recognition model and display the knowledge acquisition intent as a specific question using the text generation model.

The user portrayal database 730 is used to store basic information of the user, a core focus list, specific questions, and affinity information. The basic information may include character type, department, etc. The core point of interest list may be obtained from a point of interest calculation sub-module and the specific problem may be obtained from a problem calculation sub-module. The affinity information contains, for example, basic information of a plurality of (e.g., 50) other users most closely related to the current user.

The full set of resources 740 may be all knowledge resources that can be acquired. Including, for example, news, external articles in the internet, video, user authored articles, reports, etc. Because the quality of the knowledge resources in the corpus 740 varies, the knowledge graph construction module 750 may be used to provide insight into the knowledge resources in the corpus 740 and construct a knowledge graph.

The knowledge graph construction module 750 includes a quality operator module, a label calculation operator module, and a specialized knowledge base construction sub-module. The quality operator module is configured to perform comprehensive quality assessment on the knowledge resources in the resource corpus 740 by using the quality assessment model, filter the knowledge resources with low quality, and reserve the knowledge resources with higher quality. The label calculation sub-module is used for extracting knowledge labels of knowledge resources, the knowledge resources are used as nodes in the knowledge graph, and the knowledge labels can be used as edges of all the nodes in the knowledge graph. For example, nodes with the same knowledge labels are associated, so that knowledge resources are associated with each other to form a knowledge network. In order to recommend suitable knowledge resources to users of different character types, the expertise repository construction sub-module is configured to construct an expertise repository 760 suitable for different character types based on the knowledge network.

Professional repository 760 includes research and development professional repository, product professional repository, management professional repository, general purpose repository, and the like.

Based on the core points of interest or questions in user profile database 730, knowledge resource recall module 770 may recall a list of knowledge resources from expertise repository 760 corresponding to the user's role type. Knowledge resource recall module 770 may include a keyword recall sub-module and a semantic recall sub-module. The keyword recall sub-module is used for recalling the first knowledge resource list based on the similarity between the keywords of the knowledge resources and the core focus or problem. The semantic recall sub-module is used for recalling a second knowledge resource list according to the similarity between the semantic vector of the knowledge resource and the semantic vector of the core attention point or the problem. The knowledge resources in both the first knowledge resource list and the second knowledge resource list may participate in the ranking of the ranking module 780 as candidate knowledge resources.

The ranking module 780 includes a similarity calculation sub-module, an authority calculation sub-module, a diversity calculation sub-module, and the like, which are respectively used for calculating a similarity evaluation value between the candidate knowledge resource and the user portrait, and an authority evaluation value of the candidate knowledge resource. The ranking module 780 may also calculate a timeliness assessment value, a quality assessment value, etc., along with a similarity assessment value, an authority assessment value, and calculate a click through rate CTR for the candidate knowledge resources. And then sorting the candidate knowledge resources according to the click through rate CTR.

The diversity calculation submodule is used for calculating whether adjacent similar candidate knowledge resources exist in the ordered knowledge resource list, and only one candidate knowledge resource is reserved in the list aiming at the similar candidate knowledge resources, so that the sequence of the candidate knowledge resources is further adjusted, and a finally obtained knowledge recommendation list 790 is obtained.

Knowledge recommendation list 790 is recalled and ordered, for example, using the core focus or problem "XX Algorithm". Knowledge recommendation list 790 includes knowledge resource "XX Algorithm research" 1 st, knowledge resource "XX Algorithm optimization" 2 nd, and so on.

The whole framework of the embodiment can build a whole set of high-quality, rich and controllable knowledge recommendation products, can help enterprises to quickly build knowledge recommendation systems from 0 to 1, and solves the problem that users find pain points with difficult interested knowledge.

Fig. 8 is a block diagram of a knowledge recommendation device, according to one embodiment of the disclosure.

As shown in fig. 8, the knowledge recommendation apparatus 800 includes a first determination module 801, a first generation module 802, a second determination module 803, a recall module 804, a ranking module 805, and an output module 806.

The first determining module 801 is configured to determine knowledge requirement information of a target object according to text materials in a target object working scene.

The first generation module 802 is configured to generate a target object representation according to knowledge requirement information.

The second determining module 803 is configured to determine a target knowledge base corresponding to a role type of the target object from the plurality of specialized knowledge bases.

The recall module 804 is configured to recall the knowledge resource list from the target knowledge base according to the knowledge requirement information.

The sorting module 805 is configured to sort the knowledge resources in the knowledge resource list according to the target object representation, so as to obtain a knowledge recommendation list.

The output module 806 is configured to output the knowledge recommendation list.

According to an embodiment of the present disclosure, the text stories include work reports and the knowledge requirement information includes a core point of interest list. The first determining module comprises a first extracting unit, a screening unit, a matching unit and a sorting unit.

The first extraction unit is used for extracting a first entity word set related to the knowledge from the text material.

The screening unit is used for determining a plurality of entity words from the first entity word set as initial attention points according to the semantic quality and word frequency of the entity words in the first entity word set.

The matching unit is configured to determine at least one initial point of interest hit by the content in the work report as a candidate point of interest.

The ordering unit is used for ordering at least one candidate attention point according to the semantic similarity between the candidate attention point and the work report, and a core attention point list is obtained.

The first extraction unit is used for inputting the text material into the entity word extraction model to obtain a first entity word set.

Knowledge requirement information includes specific questions. The first determining module includes a first determining unit and a second extracting unit.

The first determining unit is used for determining target text containing knowledge acquisition intention from the text material.

The second extraction unit is used for extracting specific problems corresponding to the knowledge acquisition intention from the target text.

The first determining unit is used for inputting the text material into the intention distinguishing model to obtain a target text containing the knowledge acquisition intention.

The second extraction unit is used for inputting the target text into the problem generation model and generating a specific problem corresponding to the knowledge acquisition intention.

The first determining module comprises a second determining unit, which is used for determining a plurality of pieces of knowledge requirement information corresponding to a plurality of role types according to text materials in respective working scenes of target objects of the role types.

The knowledge recommendation device further comprises a third determination module, an association module and a division module.

The third determining module is used for determining knowledge labels of each of the plurality of knowledge resources.

The association module is used for associating knowledge resources with the same knowledge labels to obtain a knowledge network.

The division module is used for dividing the knowledge network into a plurality of professional knowledge bases corresponding to the role types according to the corresponding relation between the role types and the knowledge labels.

The knowledge tags include at least one reading tag, at least one subject tag, and a time-sensitive tag. The third determining module includes a first processing unit, a second processing unit, and a third determining unit.

The first processing unit is used for inputting the knowledge resources into the first multi-label classification model aiming at each knowledge resource to obtain at least one reading label of the knowledge resources.

The second processing unit is used for inputting the knowledge resources into the second multi-label classification model aiming at each knowledge resource to obtain at least one theme label of the knowledge resources.

The third determining unit is used for determining the timeliness label of the knowledge resource according to the creation time length of the knowledge resource and the behavior information of the knowledge resource aiming at each knowledge resource.

The knowledge recommendation device further comprises an extraction module, a screening module and a second generation module.

The extraction module is used for extracting a second entity word set related to the knowledge from each knowledge resource.

The screening module is used for determining at least one entity word from the second entity word set as a keyword of the knowledge resource according to the semantic quality and word frequency of the entity word in the second entity word set.

The second generation module is used for generating a keyword index structure of each specialized knowledge base according to the keywords of each knowledge resource in the specialized knowledge base, wherein the keyword index structure comprises a plurality of keywords and a knowledge resource identification list corresponding to each keyword.

The recall module comprises a fourth determination unit and a first recall unit.

The fourth determining unit is used for determining target keywords matched with the knowledge demand information in the keyword index structure of the target knowledge base.

The first recall unit is used for recalling the first knowledge resource list according to the knowledge resource identification list corresponding to the target keyword.

The recall module further includes a fifth determination unit, a sixth determination unit, and a second recall unit.

The fifth determining unit is used for determining semantic vectors of the knowledge requirement information.

The sixth determining unit is used for determining semantic vectors of knowledge resources in the target knowledge base.

The second recall unit is used for recalling a second knowledge resource list from the target knowledge base according to the similarity between the semantic vector of the knowledge demand information and the semantic vector of the knowledge resource.

The sorting module comprises a seventh determining unit, a first calculating unit, a second calculating unit, an eighth determining unit, a weighting unit, a ninth determining unit and an adjusting unit.

The seventh determining unit is configured to determine a knowledge resource in the first knowledge resource list and a knowledge resource in the second knowledge resource list as candidate knowledge resources;

the first calculation unit is used for calculating a correlation evaluation value between the candidate knowledge resource and the target object portrait;

the second calculation unit is used for calculating a matching degree evaluation value between the knowledge label of the candidate knowledge resource and the target object portrait;

the eighth determining unit is used for determining an authority evaluation value, a timeliness evaluation value and a quality evaluation value of the knowledge resource;

the weighting unit is used for carrying out weighting processing on the correlation evaluation value, the matching degree evaluation value, the authority evaluation value, the timeliness evaluation value and the quality evaluation value to obtain a comprehensive evaluation value;

the ninth determining unit is used for determining the ordering of the knowledge resources in the knowledge resource list according to the comprehensive evaluation value; and

The adjusting unit is used for adjusting the ordering according to the diversity requirements of the knowledge resources in the knowledge resource list to obtain a knowledge recommendation list.

The first generation module includes a tenth determination unit and a generation unit.

The tenth determination unit is configured to determine basic information and affinity information of a current target object, the affinity information including basic information of a plurality of other target objects whose relationship closeness with the current target object is greater than a threshold.

The generating unit is used for generating a target object portrait of the current target object according to the basic information, the affinity ring information and the knowledge demand information of the current target object.

According to embodiments of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

Fig. 9 shows a schematic block diagram of an example electronic device 900 that may be used to implement embodiments of the present disclosure. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the disclosure described and/or claimed herein.

As shown in fig. 9, the apparatus 900 includes a computing unit 901 that can perform various appropriate actions and processes according to a computer program stored in a Read Only Memory (ROM) 902 or a computer program loaded from a storage unit 908 into a Random Access Memory (RAM) 903. In the RAM903, various programs and data required for the operation of the device 900 can also be stored. The computing unit 901, the ROM 902, and the RAM903 are connected to each other by a bus 904. An input/output (I/O) interface 905 is also connected to the bus 904.

Various components in device 900 are connected to I/O interface 905, including: an input unit 906 such as a keyboard, a mouse, or the like; an output unit 907 such as various types of displays, speakers, and the like; a storage unit 908 such as a magnetic disk, an optical disk, or the like; and a communication unit 909 such as a network card, modem, wireless communication transceiver, or the like. The communication unit 909 allows the device 900 to exchange information/data with other devices through a computer network such as the internet and/or various telecommunications networks.

The computing unit 901 may be a variety of general and/or special purpose processing components having processing and computing capabilities. Some examples of computing unit 901 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various specialized Artificial Intelligence (AI) computing chips, various computing units running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 901 performs the respective methods and processes described above, such as the knowledge recommendation method. For example, in some embodiments, the knowledge recommendation method may be implemented as a computer software program tangibly embodied on a machine-readable medium, such as the storage unit 908. In some embodiments, part or all of the computer program may be loaded and/or installed onto the device 900 via the ROM 902 and/or the communication unit 909. When the computer program is loaded into the RAM903 and executed by the computing unit 901, one or more steps of the knowledge recommendation method described above may be performed. Alternatively, in other embodiments, the computing unit 901 may be configured to perform the knowledge recommendation method by any other suitable means (e.g. by means of firmware).

Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuit systems, field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), systems On Chip (SOCs), complex Programmable Logic Devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs, the one or more computer programs may be executed and/or interpreted on a programmable system including at least one programmable processor, which may be a special purpose or general-purpose programmable processor, that may receive data and instructions from, and transmit data and instructions to, a storage system, at least one input device, and at least one output device.

Program code for carrying out methods of the present disclosure may be written in any combination of one or more programming languages. These program code may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus such that the program code, when executed by the processor or controller, causes the functions/operations specified in the flowchart and/or block diagram to be implemented. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

To provide for interaction with a target object, the systems and techniques described here can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a target object; and a keyboard and pointing device (e.g., a mouse or a trackball) by which the target object may provide input to the computer. Other kinds of devices may also be used to provide for interaction with a target object; for example, feedback provided to the target object may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the target object may be received in any form, including acoustic input, speech input, or tactile input.

The systems and techniques described here can be implemented in a computing system that includes a background component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a target object computer having a graphical target object interface or a web browser through which a target object can interact with an implementation of the systems and techniques described here), or any combination of such background, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), wide Area Networks (WANs), and the internet.

The computer system may include a client and a server. The client and server are typically remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.

It should be appreciated that various forms of the flows shown above may be used to reorder, add, or delete steps. For example, the steps recited in the present disclosure may be performed in parallel, sequentially, or in a different order, provided that the desired results of the disclosed aspects are achieved, and are not limited herein.

The above detailed description should not be taken as limiting the scope of the present disclosure. It will be apparent to those skilled in the art that various modifications, combinations, sub-combinations and alternatives are possible, depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present disclosure are intended to be included within the scope of the present disclosure.

Claims (29)

1. A knowledge recommendation method, comprising:

determining knowledge demand information of the target object according to the text material in the target object working scene;

generating a target object image according to the knowledge demand information;

determining a target knowledge base corresponding to the role type of the target object from a plurality of professional knowledge bases;

recalling a knowledge resource list from the target knowledge base according to the knowledge demand information;

according to the target object portrait, sequencing the knowledge resources in the knowledge resource list to obtain a knowledge recommendation list; and

and outputting the knowledge recommendation list.

2. The method of claim 1, wherein the text material comprises a work report and the knowledge-demand information comprises a core point of interest list; the determining knowledge requirement information of the target object according to the text material in the target object working scene comprises the following steps:

Extracting a first entity word set related to knowledge from the text material;

determining a plurality of entity words from the first entity word set as initial attention points according to the semantic quality and word frequency of the entity words in the first entity word set;

determining at least one initial point of interest hit by content in the work report as a candidate point of interest; and

and sequencing at least one candidate attention point according to the semantic similarity between the candidate attention point and the working report to obtain the core attention point list.

3. The method of claim 2, wherein the extracting the first set of entity words related to knowledge from the text material comprises:

and inputting the text material into an entity word extraction model to obtain the first entity word set.

4. The method of claim 1, wherein the knowledge demand information includes a specific question; the determining knowledge requirement information of the target object according to the text material in the target object working scene comprises the following steps:

determining target text containing knowledge acquisition intention from the text material; and

and extracting a specific problem corresponding to the knowledge acquisition intention from the target text.

5. The method of claim 4, wherein,

the determining target text containing knowledge acquisition intention from the text material comprises:

inputting the text material into an intention judging model to obtain a target text containing knowledge acquisition intention;

the extracting the specific problem corresponding to the knowledge acquisition intention from the target text comprises the following steps:

and inputting the target text into a problem generation model to generate a specific problem corresponding to the knowledge acquisition intention.

6. The method of claim 1, wherein the determining knowledge requirement information for the target object based on text material in the target object work scene comprises:

and determining a plurality of knowledge requirement information corresponding to each of the plurality of character types according to text materials in the respective working scenes of the target objects of the plurality of character types.

7. The method of claim 1, further comprising:

determining knowledge labels of each of a plurality of knowledge resources;

correlating the knowledge resources with the same knowledge labels to obtain a knowledge network; and

and dividing the knowledge network into a plurality of professional knowledge bases corresponding to the role types according to the corresponding relation between the role types and the knowledge labels.

8. The method of claim 7, wherein the knowledge tags include at least one reading tag, at least one subject tag, and a time-lapse tag; the determining knowledge tags for each of a plurality of knowledge resources includes: for each of the knowledge resources,

inputting the knowledge resource into a first multi-label classification model to obtain at least one reading label of the knowledge resource;

inputting the knowledge resources into a second multi-label classification model to obtain at least one topic label of the knowledge resources; and

and determining the timeliness label of the knowledge resource according to the creation time of the knowledge resource and the behavior information aiming at the knowledge resource.

9. The method of claim 1, further comprising:

extracting a second entity word set related to the knowledge from the knowledge resource for each knowledge resource;

determining at least one entity word from the second entity word set as a keyword of the knowledge resource according to the semantic quality and word frequency of the entity word in the second entity word set;

and generating a keyword index structure of each specialized knowledge base according to the keywords of each knowledge resource in the specialized knowledge base aiming at each specialized knowledge base, wherein the keyword index structure comprises a plurality of keywords and a knowledge resource identification list corresponding to each keyword.

10. The method of claim 9, wherein recalling a knowledge resource list from the target knowledge base in accordance with the knowledge demand information comprises:

determining target keywords matched with the knowledge demand information in a keyword index structure of the target knowledge base;

and recalling the first knowledge resource list according to the knowledge resource identification list corresponding to the target keyword.

11. The method of claim 10, the recalling a knowledge resource list from the target knowledge base in accordance with the knowledge demand information further comprising:

determining a semantic vector of the knowledge demand information;

determining semantic vectors of knowledge resources in the target knowledge base; and

and recalling a second knowledge resource list from the target knowledge base according to the similarity between the semantic vector of the knowledge demand information and the semantic vector of the knowledge resource.

12. The method of claim 11, wherein the ranking the knowledge resources in the knowledge resource list according to the target object representation to obtain a knowledge recommendation list comprises:

determining knowledge resources in the first knowledge resource list and knowledge resources in the second knowledge resource list as candidate knowledge resources;

Calculating a correlation evaluation value between the candidate knowledge resource and the target object portrait;

calculating a matching degree evaluation value between the knowledge label of the candidate knowledge resource and the target object portrait;

determining an authority evaluation value, a timeliness evaluation value and a quality evaluation value of the knowledge resource;

weighting the correlation evaluation value, the matching degree evaluation value, the authority evaluation value, the timeliness evaluation value and the quality evaluation value to obtain a comprehensive evaluation value;

determining the ordering of the knowledge resources in the knowledge resource list according to the comprehensive evaluation value; and

and adjusting the sequencing according to the diversity requirements of the knowledge resources in the knowledge resource list to obtain the knowledge recommendation list.

13. The method of claim 1 or 12, wherein the generating a target object representation from the knowledge requirement information comprises:

determining basic information and affinity information of a current target object, wherein the affinity information comprises basic information of a plurality of other target objects with relationship closeness greater than a threshold value with the current target object; and

and generating a target object portrait of the current target object according to the basic information, the affinity ring information and the knowledge demand information of the current target object.

14. A knowledge recommendation device, comprising:

the first determining module is used for determining knowledge demand information of the target object according to the text materials in the target object working scene;

the first generation module is used for generating a target object portrait according to the knowledge demand information;

the second determining module is used for determining a target knowledge base corresponding to the role type of the target object from a plurality of professional knowledge bases;

the recall module is used for recalling a knowledge resource list from the target knowledge base according to the knowledge demand information;

the ordering module is used for ordering the knowledge resources in the knowledge resource list according to the target object portrait to obtain a knowledge recommendation list; and

and the output module is used for outputting the knowledge recommendation list.

15. The apparatus of claim 14, wherein the text material comprises a work report and the knowledge-demand information comprises a core point of interest list; the first determining module includes:

the first extraction unit is used for extracting a first entity word set related to knowledge from the text material;

the screening unit is used for determining a plurality of entity words from the first entity word set as initial attention points according to the semantic quality and word frequency of the entity words in the first entity word set;

A matching unit for determining at least one initial point of interest hit by content in the work report as a candidate point of interest; and

and the ordering unit is used for ordering at least one candidate attention point according to the semantic similarity between the candidate attention point and the work report to obtain the core attention point list.

16. The apparatus of claim 15, wherein the first extraction unit is configured to input the text material into a physical word extraction model to obtain the first physical word set.

17. The apparatus of claim 14, wherein the knowledge demand information comprises a specific question; the first determining module includes:

a first determining unit configured to determine a target text including a knowledge acquisition intention from the text material; and

and the second extraction unit is used for extracting the specific problem corresponding to the knowledge acquisition intention from the target text.

18. The apparatus of claim 17, wherein,

the first determining unit is used for inputting the text material into an intention judging model to obtain a target text containing knowledge acquisition intention;

the second extraction unit is used for inputting the target text into a problem generation model to generate a specific problem corresponding to the knowledge acquisition intention.

19. The apparatus of claim 14, wherein the first determination module comprises:

and the second determining unit is used for determining a plurality of pieces of knowledge requirement information corresponding to each of the plurality of character types according to text materials in the respective working scenes of the target objects of the plurality of character types.

20. The apparatus of claim 14, further comprising:

a third determining module, configured to determine knowledge labels of each of the plurality of knowledge resources;

the association module is used for associating knowledge resources with the same knowledge labels to obtain a knowledge network; and

the division module is used for dividing the knowledge network into a plurality of professional knowledge bases corresponding to the role types according to the corresponding relation between the role types and the knowledge labels.

21. The apparatus of claim 20, wherein the knowledge tags comprise at least one reading tag, at least one subject tag, and a time-lapse tag; the third determination module includes:

the first processing unit is used for inputting the knowledge resources into a first multi-label classification model aiming at each knowledge resource to obtain at least one reading label of the knowledge resources;

the second processing unit is used for inputting the knowledge resources into a second multi-label classification model aiming at each knowledge resource to obtain at least one theme label of the knowledge resources; and

And the third determining unit is used for determining the timeliness label of each knowledge resource according to the creation time length of the knowledge resource and the behavior information of the knowledge resource.

22. The apparatus of claim 14, further comprising:

the extraction module is used for extracting a second entity word set related to the knowledge from the knowledge resources aiming at each knowledge resource;

the screening module is used for determining at least one entity word from the second entity word set as a keyword of the knowledge resource according to the semantic quality and word frequency of the entity word in the second entity word set;

the second generation module is used for generating a keyword index structure of each specialized knowledge base according to the keywords of each knowledge resource in the specialized knowledge base, wherein the keyword index structure comprises a plurality of keywords and a knowledge resource identification list corresponding to each keyword.

23. The apparatus of claim 22, wherein the recall module comprises:

a fourth determining unit, configured to determine a target keyword that matches the knowledge requirement information in a keyword index structure of the target knowledge base;

And the first recall unit is used for recalling the first knowledge resource list according to the knowledge resource identification list corresponding to the target keyword.

24. The apparatus of claim 23, wherein the recall module further comprises:

a fifth determining unit, configured to determine a semantic vector of the knowledge requirement information;

a sixth determining unit, configured to determine a semantic vector of a knowledge resource in the target knowledge base; and

and the second recall unit is used for recalling a second knowledge resource list from the target knowledge base according to the similarity between the semantic vector of the knowledge demand information and the semantic vector of the knowledge resource.

25. The apparatus of claim 24, wherein the ranking module comprises:

a seventh determining unit, configured to determine a knowledge resource in the first knowledge resource list and a knowledge resource in the second knowledge resource list as candidate knowledge resources;

a first calculation unit configured to calculate a correlation evaluation value between the candidate knowledge resource and the target object representation;

a second calculation unit, configured to calculate a matching degree evaluation value between the knowledge tag of the candidate knowledge resource and the target object representation;

An eighth determining unit configured to determine an authority evaluation value, a timeliness evaluation value, and a quality evaluation value of the knowledge resource;

the weighting unit is used for carrying out weighting processing on the correlation evaluation value, the matching degree evaluation value, the authority evaluation value, the timeliness evaluation value and the quality evaluation value to obtain a comprehensive evaluation value;

a ninth determining unit, configured to determine, according to the comprehensive evaluation value, a ranking of knowledge resources in the knowledge resource list; and

and the adjusting unit is used for adjusting the sequencing according to the diversity requirements of the knowledge resources in the knowledge resource list to obtain the knowledge recommendation list.

26. The apparatus of claim 14 or 25, wherein the first generation module comprises:

a tenth determination unit configured to determine basic information and affinity information of a current target object, the affinity information including basic information of a plurality of other target objects whose relationship closeness with the current target object is greater than a threshold; and

and the generating unit is used for generating a target object portrait of the current target object according to the basic information, the affinity information and the knowledge requirement information of the current target object.

27. An electronic device, comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein, the liquid crystal display device comprises a liquid crystal display device,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 13.

28. A non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the method of any one of claims 1 to 13.

29. A computer program product comprising a computer program stored on at least one of a readable storage medium and an electronic device, which when executed by a processor, implements the method according to any one of claims 1 to 13.

Images