A method, device, equipment and medium for recommending think tank knowledge

Abstract

The present disclosure provides a method for recommending think tank knowledge, which includes: obtaining target user data, wherein the target user data includes: target user attributes and target user behaviors; tracking think tank knowledge according to the target user data to obtain target think tank knowledge, wherein the think tank knowledge includes at least one of the following: text, video, and sound; processing the target think tank knowledge using a personalized recommendation algorithm to obtain an initial recommendation result, wherein the initial recommendation result includes multiple initially recommended think tank knowledge; inputting the initial recommendation result into a trained recommendation model to obtain a final recommendation result, wherein the final recommendation result includes multiple finally recommended think tank knowledge. The present disclosure also provides a think tank knowledge recommendation device, equipment, storage medium, and program product.

Description

Technical Field

[0001] The present disclosure relates to the field of think tank knowledge recommendation, and specifically to a method, device, equipment, storage medium and program product for think tank knowledge recommendation. Background Art

[0002] Think tanks provide the public with a large amount of valuable information, but the amount of knowledge in think tanks is huge and the types are diverse. How to quickly track down the knowledge that users are interested in from the massive amount of think tank knowledge has become an urgent problem to be solved.

[0003] Traditional think tank knowledge tracking technologies mostly use a single tracking method based on "search". However, as think tank knowledge becomes richer, it is difficult for users to express their needs with just a few search terms. In this context, how to describe user preferences in multiple dimensions and quickly and accurately track the think tank knowledge that users are interested in from the vast amount of think tank knowledge has become an urgent problem to be solved. Summary of the invention

[0004] In view of the above problems, the present disclosure provides a method, apparatus, device, storage medium and program product for think tank knowledge recommendation.

[0005] According to a first aspect of the present disclosure, a method for recommending think tank knowledge is provided, comprising:

[0006] Obtain target user data, wherein the target user data includes: target user attributes and target user behaviors; track think tank knowledge according to the target user data to obtain target think tank knowledge, wherein the think tank knowledge includes at least one of the following: text, video, and sound; process the target think tank knowledge using a personalized recommendation algorithm to obtain an initial recommendation result, wherein the initial recommendation result includes multiple initial recommended think tank knowledge; input the initial recommendation result into a trained recommendation model to obtain a final recommendation result, wherein the final recommendation result includes multiple final recommended think tank knowledge.

[0007] According to an embodiment of the present disclosure, the recommendation model is trained in the following manner:

[0008] Obtain sample user data, wherein the sample user data includes: sample user attributes and sample user behaviors; track think tank knowledge according to the sample user data to obtain think tank knowledge samples; use a personalized recommendation algorithm to process the think tank knowledge samples to obtain initial recommendation results, wherein the initial recommendation results include multiple initial recommended think tank knowledge samples; use the sample user data and the initial recommendation results as samples in a sample pool, use a classifier to classify and predict the samples in the sample pool to obtain labeled samples, and add the labeled samples to a training set to construct a training sample set; use the training sample set to train an untrained recommendation model to obtain a trained recommendation model.

[0009] According to an embodiment of the present disclosure, the trained recommendation model is evaluated to obtain an evaluation result.

[0010] According to an embodiment of the present disclosure, a think tank knowledge sample is processed using a personalized recommendation algorithm to obtain an initial recommendation result, wherein the initial recommendation result includes a plurality of initially recommended think tank knowledge samples including:

[0011] The think tank knowledge samples are processed based on the popularity personalized recommendation algorithm to obtain the think tank knowledge samples initially recommended based on popularity; the think tank knowledge samples are processed based on the collaborative filtering personalized recommendation algorithm to obtain the think tank knowledge samples initially recommended based on collaborative filtering; the think tank knowledge samples are processed based on the matrix decomposition personalized recommendation algorithm to obtain the think tank knowledge samples initially recommended based on matrix decomposition; and the think tank knowledge samples are processed based on the content personalized recommendation algorithm to obtain the think tank knowledge samples initially recommended based on content.

[0012] According to an embodiment of the present disclosure, constructing a training sample set includes:

[0013] The threshold of the change rate of the predicted class labels of the samples in the sample pool is set to μ; when the change rate of the predicted class labels of the samples in the sample pool by the classifiers twice in a row is greater than the threshold, the uncertainty of each sample category in the sample pool is calculated; the samples with the largest uncertainty in the sample pool are labeled to obtain labeled samples; the labeled samples are removed from the sample pool and added to the training sample set to construct the training sample set; or

[0014] When the change rate of the predicted class labels of samples in the sample pool by the classifiers for two consecutive times is less than the threshold, the construction of the training sample set is completed.

[0015] A second aspect of the present disclosure provides a think tank knowledge recommendation device, comprising:

[0016] A first acquisition module is used to acquire target user data, wherein the target user data includes: target user attributes and target user behaviors;

[0017] A first tracking module is used to track think tank knowledge according to target user data to obtain target think tank knowledge, wherein the think tank knowledge includes at least one of the following: text, video, and sound;

[0018] The first recommendation module is used to process the target think tank knowledge using a personalized recommendation algorithm to obtain an initial recommendation result, wherein the initial recommendation result includes a plurality of initially recommended think tank knowledge;

[0019] The second recommendation module is used to input the initial recommendation result into the trained recommendation model to obtain the final recommendation result, wherein the final recommendation result includes multiple final recommended think tank knowledge.

[0020] According to an embodiment of the present disclosure, a training device for a think tank knowledge recommendation model includes:

[0021] A second acquisition module is used to acquire sample user data, wherein the sample user data includes: sample user attributes and sample user behaviors;

[0022] The second tracking module is used to track the think tank knowledge according to the sample user data to obtain the think tank knowledge sample, wherein the think tank knowledge includes at least one of the following: text, video, and sound;

[0023] The third recommendation module is used to process the think tank knowledge samples using a personalized recommendation algorithm to obtain an initial recommendation result, wherein the initial recommendation result includes a plurality of initially recommended think tank knowledge samples;

[0024] Construct a training sample module, which is used to use sample user data and initial recommendation results as samples in the sample pool, use a classifier to classify and predict the samples in the sample pool, obtain labeled samples, and add the labeled samples to the training set to construct a training sample set;

[0025] The training module is used to train an untrained recommendation model using a training sample set to obtain a trained recommendation model.

[0026] The third aspect of the present disclosure provides an electronic device, comprising: one or more processors; a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the above-mentioned think tank knowledge recommendation method.

[0027] The fourth aspect of the present disclosure also provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, causes the processor to execute the above-mentioned think tank knowledge recommendation method.

[0028] The fifth aspect of the present disclosure also provides a computer program product, including a computer program, which implements the above-mentioned think tank knowledge recommendation method when executed by a processor.

[0029] According to the embodiments of the present disclosure, the think tank knowledge that users are interested in can be tracked from multiple dimensions based on user attributes and user behaviors, and a larger range of think tank knowledge that users are interested in can be obtained, which is more comprehensive than the traditional single tracking method based on search. By using a personalized recommendation algorithm to make initial recommendations for a larger range of think tank knowledge that users are interested in, initial recommendation results are obtained, and the initial recommendation results are combined with deep recommendations of active learning to provide users with more accurate recommendations, solving the problem of inaccurate recommendations in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The above contents and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0031] Figure 1 A system architecture diagram schematically shows a method, apparatus, device, storage medium, and program product for recommending think tank knowledge according to an embodiment of the present disclosure;

[0032] Figure 2 A flowchart of a method for recommending think tank knowledge according to an embodiment of the present disclosure is schematically shown;

[0033] Figure 3 The schematic diagram schematically shows the framework of the method for recommending think tank knowledge according to the embodiment of the present disclosure;

[0034] Figure 4 The flowchart of the training of the recommendation model of the think tank knowledge according to the embodiment of the present disclosure is schematically shown;

[0035] Figure 5 The structure block diagram of the think tank knowledge recommendation device according to the embodiment of the present disclosure is schematically shown;

[0036] Figure 6 A structural block diagram schematically shows a training device for a think tank knowledge recommendation model according to an embodiment of the present disclosure; and

[0037] Figure 7 A block diagram of an electronic device suitable for implementing a method for think tank knowledge recommendation according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0038] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0039] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0040] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0041] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0042] In the technical solution disclosed in the present invention, the acquisition, storage and application of user personal information involved are in compliance with the provisions of relevant laws and regulations, necessary confidentiality measures are taken, and do not violate public order and good morals.

[0043] In the technical solution of the embodiment of the present disclosure, the user's authorization or consent is obtained before obtaining or collecting the user's personal information.

[0044] Think tanks are a stable social organization, referring to consulting research institutions that specialize in development research. It brings together experts and scholars from various disciplines, uses their wisdom and talents to provide satisfactory or optimized solutions for the development of social and economic fields, and is an indispensable and important part of the modern leadership and management system. Its main tasks are to provide consultation, offer advice and suggestions to decision makers, make judgments and operations, and propose various designs; feedback information, track and investigate implementation plans, and feedback the operation results to decision makers for correction. For a better understanding, take the environmental think tank as an example. The environmental think tank is composed of environmental experts and scholars, focuses on global environmental issues, and provides ideas and strategies for the country and society. It is a public research institution that shoulders social functions such as serving national environmental policies, serving knowledge dissemination, and serving social low-carbon development.

[0045] Think tank knowledge provides valuable information for everyone, which can be web page text, sound, video, etc., not limited to the information listed above. For example, an environmental protection enthusiast can browse relevant think tank websites and obtain environment-related information from them. The process of him looking for environment-related information is think tank knowledge tracking.

[0046] The following uses the target think tank knowledge and think tank knowledge samples as an example of environmental think tank knowledge, but this example is only for illustrating the target think tank knowledge and think tank knowledge samples. The target think tank knowledge and think tank knowledge samples provided by the embodiments of the present disclosure are not limited to environmental think tank knowledge.

[0047] As ecological and environmental pollution becomes increasingly serious, the public and society pay more attention to environmental protection, and people's environmental awareness is gradually improving. Environmental think tanks play an increasingly important role. Environmental think tanks provide citizens with a large amount of knowledge on sustainable development concepts, low-carbon policies, etc., and make significant contributions to the scientific nature of environmental protection. However, the amount of knowledge in think tanks is huge and varied. How to quickly track down the knowledge that users are interested in from the massive amount of think tank knowledge has become an urgent problem to be solved.

[0048] In the process of implementing the present disclosure, it is found that traditional think tank knowledge tracking technologies mostly use a single tracking method based on "search", but with the enrichment of think tank knowledge, it is difficult for users to express their needs with a few search terms. The present disclosure tracks environmental think tank knowledge based on multiple dimensions such as intelligent search, historical browsing, user attention, and personalized recommendations, and achieves accurate characterization of user preferences.

[0049] However, traditional think tank knowledge recommendation mostly adopts a single personalized recommendation technology, but the personalized recommendation technology has the problem of insufficient labeled data. In this case, an active learning model is proposed, which selects the data most needed by the model for annotation to obtain labels, and then adds the samples and their labels to the training set, so as to achieve the purpose of building a suitable training sample set at a lower cost. By training the untrained recommendation model with the constructed training set, a trained recommendation model can be obtained to solve the problem of unreliable recommendations caused by sparse data and uneven distribution in the personalized recommendation algorithm. Therefore, on the basis of the personalized recommendation algorithm, the present disclosure introduces an active learning recommendation strategy to help build a suitable training sample set, so as to obtain more reliable recommendation results.

[0050] The embodiment of the present disclosure provides a method for recommending think tank knowledge, including:

[0051] Obtain target user data, wherein the target user data includes: target user attributes and target user behaviors; track think tank knowledge according to the target user data to obtain target think tank knowledge, wherein the think tank knowledge includes at least one of the following: text, video, and sound; process the target think tank knowledge using a personalized recommendation algorithm to obtain an initial recommendation result, wherein the initial recommendation result includes multiple initial recommended think tank knowledge; input the initial recommendation result into a trained recommendation model to obtain a final recommendation result, wherein the final recommendation result includes multiple final recommended think tank knowledge.

[0052] According to the embodiments of the present disclosure, based on user attributes and user behaviors, the think tank knowledge that users are interested in can be tracked from multiple dimensions, and a larger range of think tank knowledge that users are interested in can be obtained, which is more comprehensive than the traditional single tracking method based on search. Then, a personalized recommendation algorithm is used to make initial recommendations for a larger range of think tank knowledge that users are interested in, and initial recommendation results are obtained. The initial recommendation results are combined with deep recommendations of active learning to provide users with more accurate recommendations, solving the problem of inaccurate recommendations in the prior art.

[0053] Figure 1 The system architecture diagram schematically shows the method, apparatus, device, storage medium and program product for think tank knowledge recommendation according to an embodiment of the present disclosure.

[0054] like Figure 1 As shown, the application scenario 100 according to this embodiment may include terminal devices 101, 102, a network device 104 and a server 103. The network 104 is used to provide a medium for a communication link between the terminal devices 101, 102 and the server 103. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0055] Users can use terminal devices 101 and 102 to interact with server 103 through network 104 to receive or send messages, etc. Various communication client applications and programs can be installed on terminal devices 101 and 102, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only as examples) and programs.

[0056] The terminal devices 101 and 102 may be various electronic devices having a display screen and supporting web browsing, including but not limited to a tablet computer, a laptop computer, a desktop computer, and the like.

[0057] The server 103 may be a server that provides various services, such as a background management server (only an example) that provides support for websites browsed by users using the terminal devices 101 and 102. The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device.

[0058] It should be noted that the method for recommending think tank knowledge provided in the embodiment of the present disclosure can generally be executed by the server 103. Accordingly, the apparatus for recommending think tank knowledge provided in the embodiment of the present disclosure can generally be set in the server 103. The method for recommending think tank knowledge provided in the embodiment of the present disclosure can also be executed by a server or server cluster that is different from the server 103 and can communicate with the terminal devices 101, 102 and / or the server 103. Accordingly, the apparatus for recommending think tank knowledge provided in the embodiment of the present disclosure can also be set in a server or server cluster that is different from the server 103 and can communicate with the terminal devices 101, 102 and / or the server 103.

[0059] The method for recommending think tank knowledge provided in the embodiment of the present disclosure may also be executed by the terminal devices 101 and 102, and the apparatus for recommending think tank knowledge provided in the embodiment of the present disclosure may also be generally set in the terminal devices 101 and 102. The method for recommending think tank knowledge provided in the embodiment of the present disclosure may also be executed by other terminals different from the terminal devices 101 and 102. Accordingly, the apparatus for recommending think tank knowledge provided in the embodiment of the present disclosure may also be set in other terminals different from the terminal devices 101 and 102.

[0060] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0061] The following will be based on Figure 1 The scene described by Figure 2 to Figure 4 The method for recommending think tank knowledge in the disclosed embodiment is described in detail.

[0062] Figure 2 The flowchart of the method for recommending think tank knowledge according to an embodiment of the present disclosure is schematically shown.

[0063] like Figure 2 As shown, the think tank knowledge recommendation method 200 of this embodiment includes operations S201 to S204.

[0064] In operation S201, target user data is acquired, wherein the target user data includes: target user attributes and target user behaviors.

[0065] According to the embodiments of the present disclosure, the target user attributes include: the target user's gender, age, education level, hobbies, etc. The target user behavior data can be obtained through the target user's intelligent search, historical browsing, attention and personalized recommendation, wherein the target user behavior includes: browsing, attention, sharing, downloading, etc.

[0066] According to the embodiments of the present disclosure, the intelligent search technology satisfies the functions of fuzzy search and exact matching, etc. For example, based on the Elasticsearch data analysis engine, it supports both fuzzy search with "keywords" as the search condition and precise search with multiple attributes and conditions.

[0067] In operation S202, think tank knowledge is tracked according to target user data to obtain target think tank knowledge, wherein the think tank knowledge includes at least one of the following: text, video, and sound.

[0068] According to the embodiments of the present disclosure, the table structure of the database is designed according to the target user attributes and target user behavior data, such as: designing the target user intelligent search record table, the target user browsing record table, the target user attention record table, and the target user download record table. Based on these record tables, by extracting the target user attribute characteristics and target user behavior characteristics, the target think tank knowledge is tracked and the think tank knowledge that the user may like is obtained.

[0069] In operation S203, the target think tank knowledge is processed using a personalized recommendation algorithm to obtain an initial recommendation result, wherein the initial recommendation result includes a plurality of initially recommended think tank knowledge.

[0070] According to the embodiment of the present disclosure, the target think tank knowledge is processed by using a personalized recommendation algorithm to obtain an initial recommendation result, that is, the think tank knowledge that the user may like. The personalized recommendation algorithm includes: personalized recommendation algorithms based on popularity, collaborative filtering, matrix decomposition and content.

[0071] The following uses the target think tank knowledge and think tank knowledge samples as an example of environmental think tank knowledge, but this example is only for illustrating the target think tank knowledge and think tank knowledge samples. The target think tank knowledge and think tank knowledge samples provided by the embodiments of the present disclosure are not limited to environmental think tank knowledge.

[0072] According to an embodiment of the present disclosure, the target think tank knowledge is processed using a popularity-based personalized recommendation algorithm to obtain think tank knowledge that is initially recommended based on popularity.

[0073] First, five types of environmental think tank knowledge were selected, namely, “environmental management”, “low-carbon economy”, “low-carbon city”, “sustainable development countermeasures”, and “complex environmental system”, and their initial heat values ​​were assigned. ori , assuming the initial heat value S ori=20.

[0074] As the target user interacts with a think tank's knowledge, the heat value S of the think tank's knowledge increases. user For example, for a think tank knowledge, each time it is searched, the score increases by S. sea Each time it is downloaded, the score increases by S dow Each time it is viewed, the score increases by S bro Each time you are followed, the score increases by S fol points, and then the target user behavior score can be obtained as S user =S sea +S dow +S bro +S fol .

[0075] Since think tank knowledge is highly time-sensitive, it will decay over time, which leads to the S time Therefore, according to the initial heat value of think tank knowledge, user behavior score and the decay value of think tank knowledge, the heat value of the above five types of environmental think tank knowledge can be obtained as S = S ori +S user -S time Finally, by arranging the think tank knowledge in descending order according to their popularity values, the target user can be provided with the ten most highly recommended environment-related think tank knowledge based on popularity. It should be noted that the number of recommendations is not limited to ten.

[0076] According to an embodiment of the present disclosure, the target think tank knowledge is processed using a personalized recommendation algorithm based on collaborative filtering to obtain think tank knowledge initially recommended based on collaborative filtering.

[0077] The environmental think tank knowledge is processed using the user's collaborative filtering algorithm. First, users similar to the target user are found, and then for each similar user, the environmental think tank knowledge that the similar user is interested in and has not been browsed by the target user is selected for recommendation.

[0078] The Pearson correlation coefficient method is used to calculate the similarity of users and find the five users who are most similar to the target user. The similarity of these five users is as follows:

[0079] UserU a , similarity: 0.8742; user U b , similarity: 0.8611; user U c , similarity 0.7823; user U d , similarity 0.7644; user U e , similarity 0.7422.

[0080] Then, for these five similar users, we select the environmental think tank knowledge that they are interested in and that the target user has not browsed, and calculate their scores.

[0081] For example, user U a The environmental think tank knowledge that I am most interested in is Think Tank Knowledge I a ,I b ,I c ..., the scores are 0.987, 0.982, 0.974 respectively...

[0082] Finally, by weighting the users' similarity and the scores of the environmental think tank knowledge that similar users are interested in, the target users can be provided with the ten most highly recommended environmental think tank knowledge based on collaborative filtering. It should be noted that the number of recommendations is not limited to ten.

[0083] According to an embodiment of the present disclosure, the target think tank knowledge is processed using a personalized recommendation algorithm based on matrix decomposition to obtain think tank knowledge initially recommended based on matrix decomposition.

[0084] The Baseline SVD method is used to calculate the two-dimensional rating matrix of users and environmental think tank knowledge, such as user U a , U b , U c ……, Think Tank Knowledge I a ,I b ,I c ..., using the stochastic gradient descent method to obtain a score of S aa , S ab , S ac ..., unknown scores are represented by the symbol "?". The two-dimensional score matrix is ​​shown in Table 1:

[0085] Table 1. Two-dimensional rating matrix

[0086]

[0087] The score S calculated by stochastic gradient descent method ab , S ba , S cc The scores are 0.869, 0.756, and 0.789 respectively. Sorting them in descending order according to the scores can provide users with the ten most highly recommended think tank knowledge related to the environment based on matrix decomposition. It should be noted that the number of recommendations is not limited to ten.

[0088] According to an embodiment of the present disclosure, the target think tank knowledge is processed using a content-based personalized recommendation algorithm to obtain think tank knowledge based on content initial recommendation.

[0089] First, we extract the feature information such as content, title, and type from the environmental think tank knowledge, and use the doc2vec model to encode the feature information extracted from the environmental think tank knowledge and convert the feature information into word vectors. a , realizing the feature representation of think tank knowledge, where the feature is represented by x i Indicates that the feature dimension is n.

[0090] Then, the historical behavior of the target user is used to describe the characteristic preferences of the target user. By learning the characteristics of the historical behavior of the target user, the preference vector U of the target user is obtained. b , where the feature is represented by y i Indicates that the feature dimension is n.

[0091] Finally, by calculating I a with U b The cosine similarity of the obtained cosine values ​​is arranged in descending order, and the target user can be provided with the ten most highly recommended think tank knowledge related to the environment based on content. It should be noted that the number of recommendations is not limited to ten.

[0092] In operation S204, the initial recommendation result is input into the trained recommendation model to obtain a final recommendation result, wherein the final recommendation result includes a plurality of final recommended think tank knowledge.

[0093] According to an embodiment of the present disclosure, a personalized recommendation algorithm is used to process think tank knowledge, and multiple initially recommended think tank knowledge is summarized to obtain an initial recommendation result, and the features of the initial recommendation result are vectorized as input to a trained recommendation model.

[0094] For example, the total features of the 40 think tank knowledge obtained by the initial recommendation are input into the trained recommendation model, where the total features are represented by the target user U a Knowledge with Think Tank I a composition.

[0095] Get the top 10 think tank knowledge with the highest scores as the final recommendation result, such as giving the target user U a The final recommended think tank knowledge is I a ,I b ,I c ..., the scores are 0.891, 0.895, 0.876...

[0096] Figure 3 The schematic diagram schematically shows the framework of the method for recommending think tank knowledge according to an embodiment of the present disclosure.

[0097] According to the embodiments of the present disclosure, Figure 3As shown in the figure, in order to quickly track the think tank knowledge, user data is obtained from multiple dimensions such as user attention, intelligent search, historical browsing and personalized recommendation, and the think tank knowledge that the user is interested in is tracked according to the user data. The personalized recommendation algorithm based on heat, content, collaborative filtering and matrix decomposition is used to process the think tank knowledge to obtain the initial recommendation result. The user data and the initial recommendation result are used as samples of the sample pool in the recommendation model, and the active learning algorithm strategy of uncertainty reduction is used to classify the samples in the sample pool, and the samples that do not meet the preset conditions are labeled to obtain the labeled samples. Then, the labeled samples are added to the training set as training samples, and the unlabeled samples in the sample pool are further classified and labeled using the active learning algorithm strategy of uncertainty reduction to complete the construction of the training sample set. Then, the untrained recommendation model is trained using the training samples to obtain the trained recommendation model. . In the application process, the initial recommendation results are input into the trained recommendation model to obtain more accurate recommendation results.

[0098] Through the embodiments of the present disclosure, the think tank knowledge that users are interested in is tracked from multiple dimensions according to user attributes and user behaviors, and a larger range of think tank knowledge that users are interested in is obtained, which is more comprehensive than the traditional single tracking method based on search. The personalized recommendation algorithm is used to make initial recommendations for a larger range of think tank knowledge that users are interested in, and the initial recommendation results are obtained. The initial recommendation results are combined with the deep recommendation of the active learning algorithm with uncertainty reduction, which can provide users with more accurate recommendations and solve the problem of inaccurate recommendations in the prior art.

[0099] Figure 4 The flowchart of the training of the recommendation model of the think tank knowledge according to the embodiment of the present disclosure is schematically shown.

[0100] like Figure 4 As shown, the training method 400 of the think tank knowledge recommendation model of this embodiment includes operations S401 to S405.

[0101] In operation S401, sample user data is obtained, where the sample user data includes: sample user attributes and sample user behaviors.

[0102] According to the embodiments of the present disclosure, the sample user attributes include: the sample user's gender, age, education level, hobbies, etc. The sample user behavior data can be obtained through the sample user's intelligent search, historical browsing, attention and personalized recommendation, wherein the sample user behavior includes: browsing, attention, sharing, downloading, etc.

[0103] According to the embodiments of the present disclosure, the intelligent search technology satisfies the functions of fuzzy search and exact matching, etc. For example, based on the Elasticsearch data analysis engine, it supports both fuzzy search with "keywords" as the search condition and precise search with multiple attributes and conditions.

[0104] In operation S402, think tank knowledge is tracked according to sample user data to obtain think tank knowledge samples.

[0105] According to the embodiments of the present disclosure, the table structure of the database is designed according to the sample user attributes and sample user behavior data, such as: designing a sample user intelligent search record table, a sample user browsing record table, a sample user attention record table, and a sample user download record table. Based on these sample user record tables, by extracting sample user attribute features and sample user behavior features, the think tank knowledge samples are tracked and the think tank knowledge samples that users may like are obtained.

[0106] In operation S403, the think tank knowledge samples are processed using a personalized recommendation algorithm to obtain an initial recommendation result, wherein the initial recommendation result includes a plurality of initially recommended think tank knowledge samples.

[0107] According to the embodiment of the present disclosure, the think tank knowledge samples are processed using a personalized recommendation algorithm to obtain an initial recommendation result, that is, the think tank knowledge samples that the user likes. The personalized recommendation algorithm includes: personalized recommendation algorithms based on popularity, collaborative filtering, matrix decomposition, and content.

[0108] According to an embodiment of the present disclosure, a think tank knowledge sample is processed using a popularity-based personalized recommendation algorithm to obtain a think tank knowledge sample that is initially recommended based on popularity.

[0109] First, according to the different label types of environmental think tank knowledge, each type of think tank knowledge is given a different initial heat value S ori , where the label types can be environment, economy, green travel, environmental management, low-carbon economy, low-carbon city, sustainable development countermeasures, complex environmental system, culture, society, health, safety, civilization, etc.

[0110] As sample users interact with a knowledge sample of an environmental think tank, the heat value S of the knowledge sample of the think tank increases. user For example, for a think tank knowledge sample, each time it is searched, the score increases by S. sea Each time it is downloaded, the score increases by S dow Each time it is viewed, the score increases by S bro Each time you are followed, the score increases by S fol Score, and then we can get the sample user behavior score S user =Ssea +S dow +S bro +S fol .

[0111] Since the timeliness of think tank knowledge samples is strong, they will decay over time, thus generating S time Therefore, according to the initial heat value of the think tank knowledge sample, the user behavior score and the decay value of the think tank knowledge sample, the heat value S of the think tank knowledge sample can be obtained, where S = S ori +S user -S time Finally, by sorting the think tank knowledge samples in descending order according to their popularity values, the sample users can be provided with environmental think tank knowledge samples that are initially recommended based on popularity.

[0112] According to the embodiments of the present disclosure, think tank knowledge samples are processed using a personalized recommendation algorithm based on collaborative filtering to obtain think tank knowledge samples initially recommended based on collaborative filtering.

[0113] The user's collaborative filtering algorithm is used to process the environmental think tank knowledge samples. First, users similar to the sample users are found, and then for each similar user, environmental think tank knowledge samples that the similar user is interested in and have not been browsed by the sample user are selected for recommendation.

[0114] The Pearson correlation coefficient method is used to calculate the user similarity, and the k users most similar to the sample user u are found. The similarity between the similar user v and the sample user u can be calculated using the following formula (1):

[0115]

[0116] Among them, E represents mathematical expectation and cov represents covariance.

[0117] After calculating the users similar to the sample user u through formula (1), for the sample user u, select the most similar k users and select the set of environmental think tank knowledge samples that the similar users are interested in, represented by the set S (u, k), extract the environmental think tank knowledge samples that all similar users v are interested in from S and remove the environmental think tank knowledge that the sample user u is interested in, and obtain the environmental think tank knowledge that the similar users are interested in and have not browsed by the sample user and score it. Then, the score and similarity are weighted and arranged in descending order, and then the environmental think tank knowledge samples based on the initial recommendation of collaborative filtering can be obtained. Among them, for each possible recommended environmental think tank knowledge i, the degree of interest of similar user v in it is shown in formula (2):

[0118] p(u,i)=∑ v∈S(u,k)∩N(i) w uv × vi(2)

[0119] Among them, r vi represents the degree of liking of similar user v to i, w uv Represents the similarity between sample user u and similar user v.

[0120] According to an embodiment of the present disclosure, a think tank knowledge sample is processed using a personalized recommendation algorithm based on matrix decomposition to obtain a think tank knowledge sample initially recommended based on matrix decomposition.

[0121] The Baseline SVD method is used to discover the potential factors of environmental think tank knowledge through matrix decomposition, and these factor vectors are used to describe the characteristics of sample users and environmental think tank knowledge samples.

[0122] The calculation formula of the variable value of the Baseline SVD method is as follows (3):

[0123]

[0124] where r u,i represents the score of sample user u on think tank knowledge i, u represents the global mean, b u represents the user's deviation value, b i represents the deviation value of think tank knowledge, p u,k represents the relationship between sample user u and the kth latent class, q i,k It represents the relationship between think tank knowledge i and the kth potential class. The size of K is determined by the specific number, and λ is the regularization factor.

[0125] The stochastic gradient descent method and the least squares method are used to solve the formula. Finally, by sorting in descending order according to the score S, the environmental think tank knowledge sample based on the initial recommendation of matrix decomposition can be obtained.

[0126] According to an embodiment of the present disclosure, a content-based personalized recommendation algorithm is used to process think tank knowledge samples to obtain think tank knowledge samples that are initially recommended based on content.

[0127] First, we extract the feature information of the environmental think tank knowledge, such as content, title, and type. We use the doc2vec model to encode the feature information extracted from the environmental think tank knowledge and transform the extracted feature information into word vectors to realize the feature representation of the think tank knowledge. The feature is represented by x i Indicates that the feature dimension is n.

[0128] Then, the historical behavior of the sample user is used to describe the characteristic preferences of the sample user. By learning the characteristics of the historical behavior of the sample user, the preference vector U of the sample user is obtained. b , the feature is represented by yi, and the feature dimension is n.

[0129] The cosine similarity formula is used for comparison to obtain the think tank knowledge sample based on the initial recommendation of content. The cosine similarity calculation formula is shown as follows (4):

[0130]

[0131] Finally, by calculating I a with U b The cosine similarity of is calculated, and the obtained cosine values ​​are arranged in descending order to obtain the environmental think tank knowledge sample based on content initial recommendation.

[0132] In operation S404, the sample user data and the initial recommendation results are used as samples in the sample pool, and a classifier is used to perform classification prediction on the samples in the sample pool to obtain labeled samples, and the labeled samples are added to the training set to construct a training sample set.

[0133] According to the embodiment of the present disclosure, the above-mentioned personalized recommendation algorithm based on popularity, content, collaborative filtering and matrix decomposition is used to process the think tank knowledge samples to obtain the initial recommendation results, wherein the initial recommendation results include multiple initial recommended environmental think tank knowledge samples. Then, the user data and the total data of the initial recommendation results are used as the input of the recommendation model.

[0134] According to the embodiments of the present disclosure, an active learning algorithm strategy based on uncertainty reduction is used to perform deep environmental think tank knowledge recommendation. First, the change rate of the sample class label in the sample pool is used as a measurement factor, and the classifier is used to classify and predict the samples in the sample pool, and the samples with the largest uncertainty are continuously screened out and labeled, and the labeled samples are added to the training set to construct the final training sample set.

[0135] Among them, the samples with class labels are used as training sets, and the samples without class labels are used as sample pools. The threshold of the change rate of sample prediction class labels in the sample pool is set to μ. A classifier C is trained based on the training set, and the classifier C is used to classify and predict the samples in the sample pool.

[0136] When the change rate of the predicted class labels of the samples in the sample pool by the classifiers for two consecutive times is greater than the threshold, the uncertainty of each sample category in the sample pool is calculated, and the samples with the largest uncertainty in the sample pool are labeled to obtain the labeled samples. The labeled samples are removed from the sample pool and added to the training set to construct the training sample set. The newly formed training sample set continues to train the classifier C, and then the classifier C is used to classify and predict the unlabeled samples in the sample pool until the change rate of the predicted class labels of the samples in the sample pool by the classifiers for two consecutive times is less than the threshold, then the classification prediction of the samples in the sample pool is stopped, and the construction of the training sample set is completed.

[0137] It should be noted that the method of annotating samples in the sample pool is not limited to manual annotation, but can also be machine annotation.

[0138] Through the embodiments of the present disclosure, an active learning algorithm based on uncertainty reduction is utilized, and a labeling method is adopted to select a reasonable training sample set for the recommendation model, thereby solving the technical problems caused by data sparsity and uneven distribution.

[0139] According to the embodiments of the present disclosure, XGboost is used as the classifier of the active learning algorithm. Its algorithm idea is to continuously perform feature splitting to add trees. Adding trees is to learn a new function to fit the residual of the last prediction. After the training is completed, K trees can be obtained. According to the characteristics of the sample, it can be known that the sample falls into the corresponding leaf node in each tree. The sum of the scores of the leaf nodes corresponding to each tree is the predicted value of the sample. The objective function of the XGboost algorithm is divided into two parts: the error function L(θ) and the regularization term Ω(θ). The formula is shown below (5):

[0140] obj(θ)=L(θ)+Ω(θ) (5)

[0141] The training loss function formula is shown as follows (6):

[0142]

[0143] Among them, y i is the true value, is the predicted value.

[0144] The newly generated tree needs to fit the residual value of the last prediction. After generating t trees, the prediction score can be changed to:

[0145]

[0146] At the same time, the objective function can be rewritten as:

[0147]

[0148] where g i is the first-order derivative, h i is the second-order derivative.

[0149]

[0150]

[0151] Since the residual of the prediction score of the first t-1 trees and y has no effect on the optimization of the objective function, it can be directly removed. The simplified objective function is:

[0152]

[0153] For the regularization term of the XGboost algorithm:

[0154]

[0155] The XGboost model mainly penalizes the number of leaf nodes T and the predicted value w of the leaf nodes. Penalizing the number of leaf nodes is equivalent to pruning the tree. j Represents the predicted value of the jth leaf node in a tree; T represents that a tree has a total of T leaf nodes; γ and λ are custom values ​​that can be set when using the model. If γ is large, the more leaf nodes the tree has, the greater the penalty will be, and λ will penalize the total predicted value of the leaf nodes.

[0156] Let the prediction result of the tth classifier be w q(x) :

[0157] f t (x) = w q(x) (13)

[0158] Where q(x) is the function of the leaf node, then the objective function can be changed to:

[0159]

[0160] make:

[0161]

[0162]

[0163] The final objective function can be written as:

[0164]

[0165] So get the best and obj * :

[0166]

[0167]

[0168] XGboost uses the above objective function value as the evaluation function and uses a greedy algorithm to traverse all feature partitioning points. The specific approach is that the objective function value after the split is greater than the gain value of the objective function of a single leaf node. At the same time, in order to limit the tree from growing too deep, a threshold is set. The split is only performed when the gain is greater than the threshold. At the same time, the maximum depth of the tree is set, and the growth is stopped when the sample weight is less than the set threshold to prevent overfitting. The left and right leaf nodes are distinguished by the subscripts L and R respectively. The gain score can be calculated as Gain, and the formula is as follows:

[0169]

[0170] According to the embodiment of the present disclosure, the total data of the sample user data and the initial recommendation results are divided into three parts, of which 25% is used as a training set, 25% is used as a test set, and 50% is used as samples in the sample pool for the classifier to select. In the embodiment of the present disclosure, the change rate threshold of the sample class label is set to 0.02. When the change rate of the sample category in the sample pool by the classifier for two consecutive times is less than 0.02, the construction of the training sample set is completed.

[0171] In operation S405, an untrained recommendation model is trained using the training sample set to obtain a trained recommendation model.

[0172] According to an embodiment of the present disclosure, when the number of selected samples in the sample pool is 30, the change rate of the sample category is 0.018, which is less than the set threshold of 0.02. At this time, the test accuracy of XGboost is 0.8321. When all samples in the sample pool are labeled, the test accuracy of XGboost is only 0.8123. This shows that the active learning algorithm strategy proposed in the present disclosure can achieve high-precision training effects with low-cost labeling.

[0173] According to an embodiment of the present disclosure, the trained recommendation model is evaluated to obtain an evaluation result.

[0174] According to an embodiment of the present disclosure, the initial recommended think tank knowledge is sorted based on the XGboost algorithm, and the top N environmental think tank knowledge with the highest scores are obtained as the final recommendation result.

[0175] In order to evaluate the recommendation effect, AUC and CVR are selected as evaluation indicators to measure the prediction error and the degree of user adoption of the recommendation results of the recommendation model.

[0176] The definition of AUC refers to the probability that the predicted value of a positive sample is greater than that of a negative sample when a positive sample is randomly selected from a set of positive and negative samples. The calculation formula of AUC is shown as follows (18):

[0177]

[0178] Among them, the denominator represents the total number of combinations of positive and negative samples, and the numerator refers to the number of combinations in which the predicted probability of positive samples is greater than that of negative samples.

[0179] CVR refers to conversion rate, which is an indicator to measure the effectiveness of advertising. In the embodiment of the present disclosure, when a user clicks on the recommended environmental think tank knowledge, such as the text in the environmental think tank knowledge, and browses for more than 20 seconds, the recommendation is considered to be effective. If the browsing time is too short, it means that the user is not very interested in the text. The calculation formula of CVR is as follows (19):

[0180]

[0181] Among them, effective num Refers to the number of effective clicks, total num This is the total number of clicks.

[0182] For example, AUC and CVR are used as evaluation indicators to evaluate the final recommendation effect. Experimental comparisons are conducted on single personalized recommendation algorithms such as heat-based, collaborative filtering, matrix decomposition, content-based, and the active learning-based algorithm proposed in this technology. The specific comparison results are shown in Table 2.

[0183] Table 2. Recommendation results of single personality recommendation algorithm and active learning algorithm

[0184]

[0185] It can be seen from Table 2 that the recommendation effect based on active learning algorithm is better than that of a single personalized recommendation algorithm.

[0186] Based on the above-mentioned think tank knowledge recommendation method, the present disclosure also provides a think tank knowledge recommendation device. Figure 5 The device is described in detail.

[0187] Figure 5 The structural block diagram of the think tank knowledge recommendation device according to an embodiment of the present disclosure is schematically shown.

[0188] like Figure 5 As shown, the think tank knowledge recommendation device 500 of this embodiment includes a first acquisition module 501, a first tracking module 502, a first recommendation module 503 and a second recommendation module 504.

[0189] The first acquisition module 501 is used to acquire target user data, wherein the target user data includes: target user attributes and target user behaviors. In one embodiment, the first acquisition module 501 can be used to perform the operation S201 described above, which will not be described in detail here.

[0190] The first tracking module 502 is used to track the think tank knowledge according to the target user data to obtain the target think tank knowledge, wherein the think tank knowledge includes at least one of the following: text, video, and sound. In one embodiment, the first tracking module 502 can be used to perform the operation S202 described above, which will not be repeated here.

[0191] The first recommendation module 503 is used to process the target think tank knowledge using a personalized recommendation algorithm to obtain an initial recommendation result, wherein the initial recommendation result includes a plurality of initially recommended think tank knowledge. In one embodiment, the first recommendation module 503 can be used to perform the operation S203 described above, which will not be described in detail here.

[0192] The second recommendation module 504 is used to input the initial recommendation result into the trained recommendation model to obtain the final recommendation result, wherein the final recommendation result includes a plurality of final recommended think tank knowledge. In one embodiment, the second recommendation module 504 can be used to perform the operation S204 described above, which will not be described in detail here.

[0193] According to an embodiment of the present disclosure, any multiple modules of the first acquisition module 501, the first tracking module 502, the first recommendation module 503, and the second recommendation module 504 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present disclosure, at least one of the first acquisition module 501, the first tracking module 502, the first recommendation module 503, and the second recommendation module 504 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware, and firmware, or in an appropriate combination of any of them. Alternatively, at least one of the first acquisition module 501, the first tracking module 502, the first recommendation module 503 and the second recommendation module 504 may be at least partially implemented as a computer program module, and when the computer program module is executed, the corresponding function may be executed.

[0194] Figure 6 The structural block diagram of the training device of the think tank knowledge recommendation model according to the embodiment of the present disclosure is schematically shown.

[0195] like Figure 6As shown, the training device 600 of the think tank knowledge recommendation model of this embodiment includes a second acquisition module 601, a second tracking module 602, a third recommendation module 603, a training sample construction module 604 and a training module 605.

[0196] The second acquisition module 601 is used to acquire sample user data, wherein the sample user data includes: sample user attributes and sample user behaviors. In one embodiment, the second acquisition module 601 can be used to perform the operation S401 described above, which will not be described in detail here.

[0197] The second tracking module 602 is used to track the think tank knowledge according to the sample user data to obtain a think tank knowledge sample, wherein the think tank knowledge includes at least one of the following: text, video, and sound. In one embodiment, the second tracking module 602 can be used to perform the operation S402 described above, which will not be repeated here.

[0198] The third recommendation module 603 is used to process the think tank knowledge sample using a personalized recommendation algorithm to obtain an initial recommendation result, wherein the initial recommendation result includes a plurality of initially recommended think tank knowledge samples. In one embodiment, the third recommendation module 603 can be used to perform the operation S403 described above, which will not be described in detail here.

[0199] The training sample building module 604 is used to use the sample user data and the initial recommendation results as samples in the sample pool, use the classifier to classify and predict the samples in the sample pool, obtain labeled samples, and add the labeled samples to the training set to build a training sample set. In one embodiment, the training sample building module 604 can be used to perform the operation S404 described above, which will not be repeated here.

[0200] The training module 605 is used to train an untrained recommendation model using a training sample set to obtain a trained recommendation model. In one embodiment, the training module 605 can be used to perform the operation S405 described above, which will not be repeated here.

[0201] According to an embodiment of the present disclosure, any multiple modules of the second acquisition module 601, the second tracking module 602, the third recommendation module 603, the training sample construction module 604 and the training module 605 can be combined in one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present disclosure, at least one of the second acquisition module 601, the second tracking module 602, the third recommendation module 603, the training sample construction module 604 and the training module 605 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware and firmware or in an appropriate combination of any of them. Alternatively, at least one of the second acquisition module 601, the second tracking module 602, the third recommendation module 603, the training sample construction module 604 and the training module 605 can be at least partially implemented as a computer program module, and when the computer program module is executed, the corresponding function can be executed.

[0202] Figure 7 A block diagram of an electronic device suitable for implementing a method for think tank knowledge recommendation according to an embodiment of the present disclosure is schematically shown.

[0203] like Figure 7 As shown, the electronic device 700 according to an embodiment of the present disclosure includes a processor 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage part 708 into a random access memory (RAM) 703. The processor 701 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 701 may also include an onboard memory for caching purposes. The processor 701 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0204] In RAM 703, various programs and data required for the operation of electronic device 700 are stored. Processor 701, ROM 702 and RAM 703 are connected to each other via bus 704. Processor 701 performs various operations of the method flow according to the embodiment of the present disclosure by executing the program in ROM 702 and / or RAM 703. It should be noted that the program can also be stored in one or more memories other than ROM 702 and RAM 703. Processor 701 can also perform various operations of the method flow according to the embodiment of the present disclosure by executing the program stored in the one or more memories.

[0205] According to an embodiment of the present disclosure, the electronic device 700 may further include an input / output (I / O) interface 705, which is also connected to the bus 704. The electronic device 700 may further include one or more of the following components connected to the I / O interface 705: an input portion 706 including a keyboard, a mouse, etc.; an output portion 707 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 708 including a hard disk, etc.; and a communication portion 709 including a network interface card such as a LAN card, a modem, etc. The communication portion 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the I / O interface 705 as needed. A removable medium 711, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 710 as needed, so that a computer program read therefrom is installed into the storage portion 708 as needed.

[0206] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.

[0207] According to an embodiment of the present disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, an apparatus or a device. For example, according to an embodiment of the present disclosure, a computer-readable storage medium may include the ROM 702 and / or RAM 703 described above and / or one or more memories other than ROM 702 and RAM 703.

[0208] The embodiment of the present disclosure also includes a computer program product, which includes a computer program, and the computer program contains program code for executing the method shown in the flowchart. When the computer program product is run in a computer system, the program code is used to enable the computer system to implement the method for recommending think tank knowledge provided by the embodiment of the present disclosure.

[0209] The above functions defined in the system / device of the embodiment of the present disclosure are performed when the computer program is executed by the processor 701. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0210] In one embodiment, the computer program may rely on tangible storage media such as optical storage devices, magnetic storage devices, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of signals on a network medium, and downloaded and installed through the communication part 709, and / or installed from the removable medium 711. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0211] In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 709, and / or installed from the removable medium 711. When the computer program is executed by the processor 701, the above functions defined in the system of the embodiment of the present disclosure are performed. According to the embodiment of the present disclosure, the system, device, means, module, unit, etc. described above can be implemented by a computer program module.

[0212] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect through the Internet).

[0213] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0214] It will be appreciated by those skilled in the art that the features described in the various embodiments and / or claims of the present disclosure may be combined and / or combined in a variety of ways, even if such combinations and / or combinations are not explicitly described in the present disclosure. In particular, the features described in the various embodiments and / or claims of the present disclosure may be combined and / or combined in a variety of ways without departing from the spirit and teachings of the present disclosure. All of these combinations and / or combinations fall within the scope of the present disclosure.

[0215] The embodiments of the present disclosure are described above. However, these embodiments are only for illustrative purposes and are not intended to limit the scope of the present disclosure. Although the embodiments are described above separately, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. The scope of the present disclosure is defined by the attached claims and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art may make a variety of substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A method for recommending think tank knowledge, comprising: Acquire target user data, wherein the target user data includes: target user attributes and target user behaviors; Tracking the think tank knowledge according to the target user data to obtain the target think tank knowledge, wherein the think tank knowledge includes at least one of the following: text, video, and sound; Processing the target think tank knowledge using a personalized recommendation algorithm to obtain an initial recommendation result, wherein the initial recommendation result includes a plurality of initially recommended think tank knowledge; Inputting the initial recommendation result into the trained recommendation model to obtain a final recommendation result, wherein the final recommendation result includes a plurality of final recommended think tank knowledge; The recommendation model is trained in the following way: Acquire sample user data, wherein the sample user data includes: sample user attributes and sample user behaviors; Track the think tank knowledge according to the sample user data to obtain a think tank knowledge sample; Processing the think tank knowledge samples using a personalized recommendation algorithm to obtain an initial recommendation result, wherein the initial recommendation result includes a plurality of initially recommended think tank knowledge samples; The sample user data and the initial recommendation results are used as samples in a sample pool, and a classifier is used to perform classification prediction on the samples in the sample pool to obtain labeled samples, and the labeled samples are added to a training set to construct a training sample set; Using the training sample set to train an untrained recommendation model to obtain a trained recommendation model; The think tank knowledge samples are processed by using a personalized recommendation algorithm to obtain an initial recommendation result, and the initial recommendation result includes a plurality of initially recommended think tank knowledge samples including: The think tank knowledge samples are processed based on the popularity personalized recommendation algorithm to obtain the think tank knowledge samples based on the initial popularity recommendation; The think tank knowledge samples are processed based on the collaborative filtering personalized recommendation algorithm to obtain the think tank knowledge samples based on the collaborative filtering initial recommendation; Processing the think tank knowledge samples based on the matrix decomposition personalized recommendation algorithm to obtain the think tank knowledge samples based on the initial recommendation of the matrix decomposition; and The think tank knowledge samples are processed based on the content personalized recommendation algorithm to obtain the think tank knowledge samples based on the initial content recommendation; Wherein, constructing the training sample set includes: The threshold of the change rate of the sample prediction class label in the sample pool is set to μ; When the change rate of the predicted class labels of the samples in the sample pool by the classifiers twice in a row is greater than a threshold, calculating the uncertainty of each sample category in the sample pool; Annotating the sample with the largest uncertainty in the sample pool to obtain annotated samples; The labeled samples are removed from the sample pool and added to the training sample set to construct the training sample set; or When the change rate of the predicted class labels of the samples in the sample pool by the classifiers for two consecutive times is less than a threshold, the construction of the training sample set is completed.

2. The method according to claim 1, further comprising: The trained recommendation model is evaluated to obtain an evaluation result.

3. A think tank knowledge recommendation device is used to implement the think tank knowledge recommendation method described in claim 1 or 2, and the device comprises: A first acquisition module is used to acquire target user data, wherein the target user data includes: target user attributes and target user behaviors; A first tracking module is used to track the think tank knowledge according to the target user data to obtain the target think tank knowledge, wherein the think tank knowledge includes at least one of the following: text, video, and sound; A first recommendation module is used to process the target think tank knowledge using a personalized recommendation algorithm to obtain an initial recommendation result, wherein the initial recommendation result includes a plurality of initially recommended think tank knowledge; The second recommendation module is used to input the initial recommendation result into the trained recommendation model to obtain the final recommendation result, wherein the final recommendation result includes multiple final recommended think tank knowledge.

4. The device according to claim 3, wherein: The training device of the recommendation model includes: A second acquisition module is used to acquire sample user data, wherein the sample user data includes: sample user attributes and sample user behaviors; A second tracking module is used to track the think tank knowledge according to the sample user data to obtain a think tank knowledge sample, wherein the think tank knowledge includes at least one of the following: text, video, and sound; A third recommendation module is used to process the think tank knowledge sample using a personalized recommendation algorithm to obtain an initial recommendation result, wherein the initial recommendation result includes a plurality of initially recommended think tank knowledge samples; Constructing a training sample module, which is used to use the sample user data and the initial recommendation results as samples in a sample pool, use a classifier to perform classification prediction on the samples in the sample pool to obtain labeled samples, and add the labeled samples to a training set to construct a training sample set; The training module is used to train an untrained recommendation model using the training sample set to obtain a trained recommendation model.

5. An electronic device comprising: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors execute the method according to any one of claims 1 to 2.

6. A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, causes the processor to execute the method according to any one of claims 1 to 2.

7. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method according to any one of claims 1 to 2 is implemented.

Images