Content recommendation method and device, electronic equipment and storage medium

By obtaining user logs and calculating weights using multiple recall strategies and prediction models, the problem that content recommendation in the existing technology is difficult to meet the personalized needs of scientific researchers, personalized content recommendation is realized, and the research field refinement and Changeable demand.

CN118939784APending Publication Date: 2024-11-12BEIJING GRIDSUM TECH CO LTD

Patent Information

Application Number
CN202310532842.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-11
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The existing content recommendation strategy is difficult to meet the personalized needs of scientific researchers, especially when the research field is refined and the research direction is changing, and the fusion weight of user dynamic data and static data is difficult to design, so it is impossible to effectively deal with users. personalized questions.

Method used

By obtaining the logs of users to be recommended, using multiple recall strategies to recall the logs, using prediction models to calculate the weights of each recall strategy, recommend content based on the weights, and adjust in real time to meet user needs. Specific steps include obtaining logs, performing a recall, calculating the average click-through rate of the recall results to determine the weight of the recall strategy, and providing content recommendations based on these weights.

Benefits of technology

It realizes personalized content recommendations, can effectively meet user needs, and provides more accurate literature recommendations for scientific researchers' research fields and changing research directions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118939784A_ABST
    Figure CN118939784A_ABST
Patent Text Reader

Abstract

The embodiment of the invention provides a content recommendation method and device, electronic equipment and a storage medium, and belongs to the field of information processing. The method comprises the following steps: acquiring a log of a to-be-recommended user; recalling the log according to a plurality of recall strategies to obtain a plurality of corresponding recall results; respectively substituting the recall results into a prediction model to obtain weights of corresponding recall strategies; and performing content recommendation according to the weight of the recall strategy. According to the method, the recommendation strategy is personalized, and user requirements are effectively met.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of information processing, and in particular to a method, device, electronic device and storage medium for content recommendation. Background Art

[0002] With the development of society and scientific progress, the massive amount of electronic documents not only provides basis and help for scientific researchers, but also increases the difficulty of overall search. In addition, due to the scarcity of user behavior data, the complexity of document types, and the difficulty in distinguishing content, the existing recommendation strategies for documents are difficult to meet user needs.

[0003] In order to solve the above problems, a method based on combining user dynamic data (such as user click and search data) and static data (such as data on the user's department) is usually adopted to make up for the lack of dynamic data. However, due to the characteristics of user static data such as wide range and slow update, the existing technology cannot cope with the problems of scientific researchers' research field refinement and changing research directions. The recommendation strategy based on the combination of user dynamic data and static data is also artificially designed based on expert opinions, which makes it difficult to solve problems such as personalization. Summary of the invention

[0004] The purpose of the embodiments of the present invention is to provide a method, device, electronic device and storage medium for content recommendation, which personalizes the recommendation strategy and effectively meets the needs of users.

[0005] In order to achieve the above-mentioned purpose, an embodiment of the present invention provides a method for content recommendation, which includes: obtaining a log of a user to be recommended; recalling the log according to multiple recall strategies to obtain corresponding multiple recall results; respectively bringing the recall results into a prediction model to obtain the weights of the corresponding recall strategies; and recommending content according to the weights of the recall strategies.

[0006] Optionally, the step of respectively bringing the recall results into a prediction model to obtain weights of corresponding recall strategies comprises: respectively bringing the recall results into a prediction model to obtain average click-through rates of corresponding recall strategies; and determining the weights of the recall strategies according to the average click-through rates of the recall strategies.

[0007] Optionally, the step of bringing the recall result into a prediction model to obtain an average click rate of a recall strategy corresponding to the recall result includes:

[0008]

[0009] Among them, x is the recall result, n is the number of documents corresponding to the recall strategy, θ is the model weight, and CTR is the average click rate of the recall strategy.

[0010] Optionally, determining the weight of the recall strategy according to the average click rate of the recall strategy includes:

[0011]

[0012] Among them, w i is the weight of the recall strategy i, is the average click rate of recall strategy i in all positions, and F() is the reverse prediction model.

[0013] Optionally, the log includes at least one of a retrieval log, a click log and a browsing log; the recall strategy is at least one of a search term recall, a dynamic portrait similarity recall, a dynamic portrait hot recall and a similarity recall of read documents.

[0014] Optionally, the prediction model is trained using the log.

[0015] On the other hand, the present invention provides a device for content recommendation, which includes: an acquisition device for acquiring a log of a user to be recommended; a recall device for recalling the log according to a plurality of recall strategies to obtain a corresponding plurality of recall results; a processing device for respectively bringing the recall results into a prediction model to obtain the weights of the corresponding recall strategies, and performing document recommendation according to the weights of the recall strategies.

[0016] Optionally, the step of respectively bringing the recall results into a prediction model to obtain weights of corresponding recall strategies comprises: respectively bringing the recall results into a prediction model to obtain average click-through rates of corresponding recall strategies; and determining the weights of the recall strategies according to the average click-through rates of the recall strategies.

[0017] On the other hand, the present invention provides an electronic device, comprising: at least one processor; and a memory connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the at least one processor implements the above-mentioned content recommendation method by executing the instructions stored in the memory.

[0018] On the other hand, the present invention provides a machine-readable storage medium storing machine instructions, which, when executed on a machine, enable the machine to execute the above-mentioned content recommendation method.

[0019] The content recommendation method of the present invention comprises: obtaining the log of the user to be recommended; recalling the log according to multiple recall strategies to obtain multiple corresponding recall results; respectively bringing the recall results into the prediction model to obtain the weight of the corresponding recall strategy; and recommending content according to the weight of the recall strategy. The method adopts the training and learning of the weight allocation of multiple recommendation strategies to personalize the recommendation strategy, and at the same time adjusts the weight in real time according to the user's click behavior to meet the user's needs.

[0020] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0022] Figure 1 is a flow chart of a content recommendation method of the present invention;

[0023] Figure 2 It is a schematic diagram of an electronic device of the present invention. DETAILED DESCRIPTION

[0024] The specific implementation of the embodiment of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the embodiment of the present invention, and is not used to limit the embodiment of the present invention.

[0025] Figure 1 is a flow chart of a content recommendation method of the present invention, such as Figure 1 As shown, step S101 is to obtain the log of the user to be recommended. The log includes at least one of a search log, a click log and a browsing log. The content recommendation includes a document recommendation and the like.

[0026] Step S102 is to recall the log according to a plurality of recall strategies to obtain a plurality of corresponding recall results. The recall strategy is at least one of search term recall, dynamic portrait similarity recall, dynamic portrait popular recall and similarity recall of read documents. The recall refers to triggering as many correct results as possible from the full information set, and returning the results to the sorting as input. The recall strategy can be simply divided into functional recall, popular recall, business recall and personalized recall, etc. Personalized recall can effectively grasp the user's interest preferences and occupies a very important position in the entire recall system. The recall strategy of the present application is to make recommendations based on the personalization of the user to be recommended.

[0027] According to a specific implementation, the present invention uses word segmentation and other methods to convert the keywords, abstracts and other features of the document into the label of the document. Search term recall uses the similarity between the user's search terms and the document label to recall documents similar to the user's search record, and the similarity calculation is based on the TF-IDF algorithm. The TF-IDF (termfrequency-inverse document frequency) is a commonly used weighted technology for information retrieval and text mining. TF-IDF is a statistical method for evaluating the importance of a word to a document set or one of the documents in a corpus. The importance of a word increases in direct proportion to the number of times it appears in the document, but at the same time it decreases in inverse proportion to the frequency of its appearance in the corpus. The main idea of ​​TF-IDF is: if a word appears in an article with a high frequency TF and rarely appears in other articles, it is considered that this word or phrase has a good ability to distinguish categories and is suitable for classification.

[0028] The user's dynamic portrait recall uses the document tags and the user's click rate on the document to build the user's dynamic portrait. The similarity recall of the dynamic portrait calculates the similarity between the user's dynamic portrait and the document tag, and then realizes the recall. The popular recall of the dynamic portrait recalls the popular documents related to the user based on the popularity of the document and the user's dynamic portrait.

[0029] The recall strategies all utilize the label features of the document, while the similarity recall of the read documents and other strategies utilize the vector features of the document. The features of different categories of documents (such as titles and abstracts) are converted into vectors using a deep learning model. The present invention implements the vectorization of text features based on Word2vec. The vector features are then used to realize the similarity recall of the documents that the user has read. The Word2vec described herein is a form of converting words into corresponding vectors. Modeling in Word2vec is not the ultimate goal, but the purpose is to obtain modeling parameters.

[0030] Step S103 is to bring the recall results into the prediction model to obtain the weight of the corresponding recall strategy. The step of bringing the recall results into the prediction model to obtain the weight of the corresponding recall strategy includes: bringing the recall results into the prediction model to obtain the average click-through rate of the corresponding recall strategy; and determining the weight of the recall strategy according to the average click-through rate of the recall strategy.

[0031] After each recall scheme obtains the recall results, experts are used to manually design the initial weight distribution and sorting scheme.

[0032] According to a specific implementation method, the fusion ratio of multiple recall results is trained and adjusted in real time according to the user's click data. The recall result is brought into the prediction model to obtain the average click rate of the recall strategy corresponding to the recall result, including: Among them, x is the recall result, n is the number of documents corresponding to the recall strategy, θ is the model weight, and CTR is the average click rate of the recall strategy.

[0033] Determining the weight of the recall strategy according to the average click rate of the recall strategy includes: Among them, w i is the weight of the recall strategy i, is the average click rate of recall strategy i in all positions, and F() is the reverse prediction model. The two key issues of recommendation strategy are recall and ranking. The recall refers to triggering as many correct results as possible from the full information set and returning the results to the ranking.

[0034] According to the above model, the model is trained offline at fixed intervals using the user's click data to adjust the fusion ratio of the multi-channel recall results. The recall strategy adopted by the present invention makes full use of the user's dynamic data and uses a variety of recommendation strategies to solve the problem of the refinement of literature research fields and the variability of user research fields.

[0035] Step S104 is to recommend content according to the weight of the recall strategy. The content recommendation is given priority if the weight of the recall strategy is prioritized. According to the user's click behavior, the weight is adjusted in real time to meet the user's needs. The present invention proposes a multi-channel fusion recommendation strategy based on user dynamic data. This strategy makes full use of the user's dynamic data and uses a variety of recommendation strategies to solve the problems of the refinement of literature research fields and the variability of user research fields. At the same time, a model is added to realize the training and learning of the weight allocation of multiple recommendation strategies, and the recommendation strategy is personalized. At the same time, the weight can be adjusted in real time according to the user's click behavior to meet the user's needs.

[0036] The present invention also provides a device for content recommendation, which includes: an acquisition device for acquiring a log of a user to be recommended; a recall device for recalling the log according to a plurality of recall strategies to obtain a plurality of corresponding recall results; a processing device for respectively bringing the recall results into a prediction model to obtain the weights of the corresponding recall strategies, and recommending documents according to the weights of the recall strategies. The respectively bringing the recall results into the prediction model to obtain the weights of the corresponding recall strategies includes: respectively bringing the recall results into the prediction model to obtain the average click-through rate of the corresponding recall strategies; and determining the weight of the recall strategies according to the average click-through rate of the recall strategies. The device and the method personalize the recommendation strategies and effectively meet the needs of users.

[0037] A content recommendation method of the present invention includes: obtaining a log of a user to be recommended; recalling the log according to multiple recall strategies to obtain multiple corresponding recall results; respectively bringing the recall results into a prediction model to obtain the weight of the corresponding recall strategy; and recommending content according to the weight of the recall strategy. The method uses training and learning to allocate weights for multiple recommendation strategies to personalize the recommendation strategy, and at the same time, adjusts the weight in real time according to the user's click behavior to meet the user's needs.

[0038] Figure 2 is a schematic diagram of an electronic device of the present invention, such as Figure 2 As shown, the electronic device includes: at least one processor; a memory connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the at least one processor implements the content recommendation method by executing the instructions stored in the memory.

[0039] The processor contains a kernel, which calls the corresponding program unit from the memory. One or more kernels can be set, and the kernel parameters can be adjusted to adopt a variety of recommendation strategies to allocate weights for training and learning, giving the recommendation strategy personalization. At the same time, the weights are adjusted in real time according to the user's click behavior to meet user needs.

[0040] An embodiment of the present invention provides a storage medium on which a program is stored. When the program is executed by a processor, the content recommendation method is implemented.

[0041] An embodiment of the present invention provides a processor, which is used to run a program, wherein the method for content recommendation is executed when the program is running.

[0042] The embodiment of the present invention provides a device, which includes at least one processor, at least one memory connected to the processor, and a bus; wherein the processor and the memory communicate with each other through the bus; the processor is used to call program instructions in the memory to execute the above-mentioned content recommendation method. The device in this article can be a server, a PC, a PAD, a mobile phone, etc.

[0043] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program that initializes the following method steps: obtaining the log of the user to be recommended; recalling the log according to multiple recall strategies to obtain corresponding multiple recall results; respectively bringing the recall results into the prediction model to obtain the weights of the corresponding recall strategies; and recommending content according to the weights of the recall strategies.

[0044] The step of bringing the recall results into the prediction model to obtain the weights of the corresponding recall strategies includes: bringing the recall results into the prediction model to obtain the average click-through rate of the corresponding recall strategies; and determining the weights of the recall strategies according to the average click-through rate of the recall strategies. The step of bringing the recall results into the prediction model to obtain the average click-through rate of the recall strategies corresponding to the recall results includes:

[0045]

[0046] Wherein, x is the recall result, n is the number of documents corresponding to the recall strategy, θ is the model weight, and CTR is the average click-through rate of the recall strategy. Determining the weight of the recall strategy according to the average click-through rate of the recall strategy includes:

[0047]

[0048] Among them, w i is the weight of the recall strategy i, is the average click rate of recall strategy i in all positions, and F() is the reverse prediction model. The log includes at least one of retrieval log, click log and browsing log; the recall strategy is at least one of search term recall, dynamic portrait similarity recall, dynamic portrait hot recall and similarity recall of read documents.

[0049] The present invention also proposes training the prediction model through the log, such as using the user's click data to perform offline training on the model at fixed intervals.

[0050] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0051] In a typical configuration, the device includes one or more processors (CPU), memory and bus. The device may also include input / output interface, network interface and the like.

[0052] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip. The memory is an example of a computer-readable medium.

[0053] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0054] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0055] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0056] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A method for content recommendation, It is characterized in that The method includes: Get the log of the user to be recommended; Recalling the logs according to multiple recall strategies to obtain corresponding multiple recall results; The recall results are respectively brought into the prediction model to obtain the weights of the corresponding recall strategies; Content recommendation is performed according to the weight of the recall strategy.

2. The method according to claim 1, It is characterized in that The step of bringing the recall results into the prediction model to obtain the weights of the corresponding recall strategies includes: The recall results are respectively brought into the prediction model to obtain the average click rate of the corresponding recall strategy; The weight of the recall strategy is determined according to the average click rate of the recall strategy.

3. The method according to claim 2, It is characterized in that The step of bringing the recall result into the prediction model to obtain the average click rate of the recall strategy corresponding to the recall result includes: Wherein, x is the recall result, n is the number of documents corresponding to the recall strategy, θ is the model weight, CTR is the average click-through rate of the recall strategy.

4. The method according to claim 2, It is characterized in that Determining the weight of the recall strategy according to the average click rate of the recall strategy includes: Among them, w i is the weight of the recall strategy i, is the average click rate of recall strategy i in all positions, F() is the reverse prediction model.

5. The method according to claim 1, It is characterized in that The log includes at least one of a search log, a click log, and a browsing log; The recall strategy is at least one of search term recall, dynamic portrait similarity recall, dynamic portrait hot recall and similarity recall of read documents.

6. The method according to claim 1, It is characterized in that The prediction model is trained using the logs.

7. A device for content recommendation, It is characterized in that The device includes: An acquisition device, used to acquire the log of the user to be recommended; A recall device, used to recall the log according to multiple recall strategies to obtain corresponding multiple recall results; The processing device is used to bring the recall results into the prediction model to obtain the weights of the corresponding recall strategies, and to make document recommendations based on the weights of the recall strategies.

8. The device according to claim 7, It is characterized in that The step of bringing the recall results into the prediction model to obtain the weights of the corresponding recall strategies includes: The recall results are respectively brought into the prediction model to obtain the average click rate of the corresponding recall strategy; The weight of the recall strategy is determined according to the average click rate of the recall strategy.

9. An electronic device, It is characterized in that The electronic device includes: at least one processor; a memory connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the at least one processor implements the content recommendation method described in any one of claims 1 to 6 by executing the instructions stored in the memory.

10. A machine-readable storage medium, It is characterized in that Machine instructions are stored, and when the machine instructions are executed on a machine, the machine executes the content recommendation method described in any one of claims 1 to 6.

Citation Information

Patent Citations

  • Recommendation method and device, computer readable storage medium and electronic equipment

    CN112115370A

  • Information pushing method, device and equipment and storage medium

    CN112131456A

  • Weight adjustment method and device and storage medium

    CN112836085A

  • Method and device for determining recommended resources, electronic equipment and storage medium

    CN113326436A

  • Personalized information recommendation method and device, equipment and storage medium

    CN115618121A

Cited By

  • Large language model knowledge base question answering system based on multi-path fusion recall retrieval algorithm

    CN120144773A