A topic selection method, device and storage medium based on artificial intelligence

Abstract

The present application discloses a topic selection method, device and storage medium based on artificial intelligence, the method comprising: obtaining the subject words input by the target object; determining the technical fields of interest to the target object based on the historical reading data of the target object; determining multiple sub-topics that meet the topic selection needs of the target object based on the technical fields and subject words of interest to the target object; performing multi-dimensional analysis on each sub-topic, and recommending the multiple sub-topics to the target object after sorting them according to the results of the analysis. The present application achieves the purpose of personalized and accurate recommendation of sub-topics for users, and achieves the technical effect of ensuring that the sub-topics recommended by the topic selection platform are highly matched with the user's topic selection needs. This solves the technical problem in the prior art that the existing topic recommendation method lacks a personalized recommendation mechanism, resulting in the recommended topics not matching the researcher's topic selection needs.

Description

Technical Field

[0001] The present application relates to the technical field of topic selection, and in particular to an artificial intelligence-based topic selection method, device and storage medium. Background Art

[0002] In the era of information explosion, the topic selection of scientific research faces unprecedented challenges. The traditional method of topic selection, whether relying on personal experience and intuition or simply browsing literature, is difficult to effectively cope with the impact of massive information and the rapidly changing scientific research environment. Especially for those researchers who hope to explore new knowledge in interdisciplinary fields, it is becoming increasingly difficult to find a topic that has both depth and breadth. The traditional literature review method is not only time-consuming and laborious, but also often fails to capture the latest research trends and potential scientific research opportunities.

[0003] Although some existing topic selection tools can help researchers identify hot research areas and predict future research trends by analyzing large-scale scientific literature databases, these topic selection tools are still at the basic stage of data search and organization, and fail to deeply understand the complexity and diversity of research topics. In addition, they fail to fully consider the unique background and interests of each researcher, lack a personalized recommendation mechanism, and the recommended topics cannot match the researcher's topic selection needs.

[0004] Currently, no effective solution has been proposed to the problem that the existing topic recommendation method lacks a personalized recommendation mechanism, resulting in the recommended topics not matching the researcher's topic needs. Summary of the invention

[0005] The embodiments of the present disclosure provide a topic selection method, device and storage medium based on artificial intelligence, so as to at least solve the technical problem that the existing topic recommendation method in the prior art lacks a personalized recommendation mechanism, resulting in the recommended topic not matching the topic selection requirements of the researcher.

[0006] According to one aspect of the embodiments of the present disclosure, there is provided a topic selection method based on artificial intelligence, including: obtaining a subject word input by a target object; determining the technical field of interest to the target object based on the historical reading data of the target object; determining multiple sub-topics that meet the target object's topic selection needs based on the technical field and subject words of interest to the target object; performing a multi-dimensional analysis on each sub-topic, and recommending the multiple sub-topics to the target object after sorting them according to the results of the analysis. According to another aspect of the embodiments of the present disclosure, there is also provided a storage medium, the storage medium including a stored program, wherein the above method is executed by a processor when the program is running.

[0007] According to another aspect of the embodiment of the present disclosure, there is also provided a topic selection device based on artificial intelligence, including: an acquisition module, used to acquire keywords input by a target object; a first determination module, used to determine the technical field of interest to the target object based on the historical reading data of the target object; a second determination module, used to determine multiple sub-topics that meet the target object's topic selection requirements based on the technical fields and keywords of interest to the target object; an analysis and recommendation module, used to perform multi-dimensional analysis on each sub-topic, and sort the multiple sub-topics according to the results of the analysis and recommend them to the target object.

[0008] According to another aspect of the embodiment of the present disclosure, there is also provided a topic selection device based on artificial intelligence, including a processor; and a memory connected to the processor, for providing the processor with instructions for processing the following processing steps: obtaining a subject word input by a target object; determining the technical field of interest to the target object based on the historical reading data of the target object; determining a plurality of sub-topics that meet the topic selection needs of the target object based on the technical fields and subject words of interest to the target object; performing a multi-dimensional analysis on each sub-topic, and ranking the plurality of sub-topics according to the results of the analysis and recommending them to the target object.

[0009] Thus, when a user has a topic selection requirement, he can input the corresponding keyword into his terminal device, and the topic selection platform can obtain the keyword input by the user from the terminal device. Then, the topic selection platform determines the technical field of the user's concern based on the user's historical reading data, laying the foundation for the subsequent personalized and accurate recommendation of sub-topics. Secondly, the topic selection platform comprehensively considers the technical field of the user's concern and the input keyword, and determines multiple sub-topics that meet the user's topic selection requirements, thereby realizing personalized sub-topic customization services and providing data support for subsequent accurate topic recommendations. Finally, the topic selection platform analyzes each sub-topic in multiple dimensions, and recommends multiple sub-topics to users after accurately sorting them according to the results of the analysis, thereby achieving the purpose of personalized and accurate recommendation of sub-topics for users, and achieving the technical effect of ensuring that the sub-topics recommended by the topic selection platform are highly matched with the user's topic selection requirements. This solves the technical problem that the existing topic recommendation method in the prior art cannot match the researcher's topic selection requirements due to the lack of a personalized recommendation mechanism. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The drawings described herein are used to provide a further understanding of the present disclosure and constitute a part of the present application. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation on the present disclosure. In the drawings:

[0011] Figure 1 is a hardware structure block diagram of a computing device for implementing the method according to Embodiment 1 of the present application;

[0012] Figure 2 is a schematic diagram of the topic selection platform based on artificial intelligence according to Example 1 of the present application;

[0013] Figure 3 It is a flowchart of the topic selection method based on artificial intelligence according to Example 1 of the present application;

[0014] Figure 4 is a schematic diagram of a process for determining a comprehensive feature vector according to Embodiment 1 of the present application;

[0015] Figure 5 is a schematic diagram of a process for determining the probability value of selecting the j-th sub-topic according to Embodiment 1 of the present application;

[0016] Figure 6 is a schematic diagram of the topic selection device based on artificial intelligence according to Example 2 of the present application; and

[0017] Figure 7 It is a schematic diagram of the topic selection device based on artificial intelligence described in Example 3 of the present application. DETAILED DESCRIPTION

[0018] In order to enable those skilled in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only embodiments of a part of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present disclosure.

[0019] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices.

[0020] Example 1

[0021] According to this embodiment, a method embodiment of a topic selection method based on artificial intelligence is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0022] The method embodiment provided in this embodiment can be executed in a mobile terminal, a computer terminal, a server or a similar computing device. Figure 1 FIG. 1 shows a hardware structure block diagram of a computing device for implementing a topic selection method based on artificial intelligence. Figure 1 As shown, the computing device may include one or more processors (the processor may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory for storing data, a transmission device for communication functions, and an input / output interface. The memory, the transmission device, and the input / output interface are connected to the processor via a bus. In addition, it may also include: a display, a keyboard, and a cursor control device connected to the input / output interface. A person skilled in the art can understand that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0023] It should be noted that the one or more processors and / or other data processing circuits described above may generally be referred to herein as "data processing circuitry". The data processing circuitry may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuitry may be a single independent processing module, or may be incorporated in whole or in part into any of the other components of the computing device. As involved in the embodiments of the present disclosure, the data processing circuitry acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0024] The memory can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the topic selection method based on artificial intelligence in the embodiment of the present disclosure. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, the topic selection method based on artificial intelligence of the above-mentioned application is realized. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the computing device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0025] The transmission device is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computing device. In one example, the transmission device includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device can be a radio frequency (Radio Frequency, RF) module, which is used to communicate with the Internet wirelessly.

[0026] The display may be, for example, a touch screen liquid crystal display (LCD) that may enable a user to interact with a user interface of the computing device.

[0027] It should be noted that, in some optional embodiments, the above Figure 1 The computing device shown may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of both hardware elements and software elements. Figure 1 This is merely one example of a particular embodiment and is intended to illustrate the types of components that may be present in the computing devices described above.

[0028] Figure 2 is a schematic diagram of the topic selection platform based on artificial intelligence according to this embodiment. Figure 2As shown, the system includes: terminal devices 110-130 of each user 310-330 with topic selection needs and a topic selection platform 200. When any user (for example but not limited to user 310) has a topic selection need, the corresponding keyword can be sent to the topic selection platform 200 through the terminal device 110. The topic selection platform 200 determines multiple sub-topics that meet the topic selection needs of user 310 based on the keyword and returns them to the terminal device 110, which then performs a visual display to the user 310. The user 310 can determine his own topic by comparing the various sub-topics, thereby achieving personalized and accurate recommendations for the user.

[0029] The topic selection platform 200 provides functions such as document query, knowledge push and communication community. In the document query function, users can query corresponding documents (such as but not limited to patent texts, academic papers, journals, blog articles, research reports and forum posts, etc.) according to their own needs, and support users to view and annotate the queried documents online. In the knowledge push function, based on user portrait technology, the user's scientific research direction, scientific research interests, access behavior and other information can be tracked to achieve personalized, accurate and intelligent knowledge push for users, and support users to view and annotate the pushed knowledge online. In the communication community function, popular posts, popular resources, notices, community active information, etc. are provided to guide scientific researchers to participate in exchanges, and support users to view and annotate all information in the communication community online.

[0030] In the above operating environment, according to the first aspect of this embodiment, a topic selection method based on artificial intelligence is provided. The method comprises Figure 2 The topic selection platform 200 shown in is implemented. Figure 3 A schematic diagram showing the process of the method is shown in FIG. Figure 3 As shown, the method includes:

[0031] S102: Obtaining a subject word input by a target object;

[0032] S104: Determine the technical field that the target object is interested in based on the historical reading data of the target object;

[0033] S106: Determine multiple sub-topics that meet the topic selection needs of the target object based on the technical fields and keywords that the target object is interested in;

[0034] S108: Analyze each sub-topic in multiple dimensions, and rank the multiple sub-topics according to the analysis results and recommend them to the target object.

[0035] Specifically, refer to Figure 2As shown, the target object is, for example, but not limited to, user 310, user 320, or user 330. When user 310, user 320, or user 330 has a topic selection requirement, they can log in to the topic selection platform 200 through their respective terminal devices, and then input a keyword corresponding to the topic selection requirement, and the corresponding terminal device sends the keyword to the topic selection platform 200. Thus, the topic selection platform 200 can obtain the keyword input by the target object from the terminal device 110 (S102). Among them, the way to log in to the topic selection platform 200 is, for example, but not limited to, any one of WeChat login, mobile phone number login, email login, platform account login, etc.

[0036] In this embodiment, the method described in this embodiment is explained by taking the target object as user 310, and the topic selection requirement of user 310 is the research on new sustainable energy technology, and the keyword input by the user 310 is "sustainable energy". Those skilled in the art should understand that the method described in this embodiment is also applicable to providing personalized topic recommendations for other users.

[0037] After the topic selection platform 200 obtains the keyword "sustainable energy" input by the user 310 from the terminal device 110, considering that the reading behavior data of the user 310 can reflect the scientific research focus and interest of the user 310 to a certain extent, the topic selection platform 200 can collect the historical reading data of the user 310, and then determine the technical field that the user 310 is concerned about based on the historical reading data of the user 310, laying the foundation for the subsequent personalized and accurate recommendation of sub-topics (S104). Specifically, the topic selection platform 200 analyzes the collected historical reading data of the user 310 and determines that the technical fields that the user 310 is concerned about include, for example but not limited to: solar energy technology, wind energy technology, hydropower energy technology, biomass energy technology, geothermal energy technology, tidal energy technology, and nuclear energy technology.

[0038] Then, the topic selection platform 200 determines multiple sub-topics that meet the topic selection needs of the user 310 based on the technical fields that the user 310 is concerned about and the keywords that the user 310 inputs (S106), thereby realizing personalized sub-topic customization services and providing data support for subsequent precise topic recommendations. Specifically, the topic selection platform 200 comprehensively considers the technical fields that the user 310 is concerned about and the keywords that the user 310 inputs, that is, comprehensively considers solar energy technology, wind energy technology, hydropower technology, biomass energy technology, geothermal energy technology, tidal energy technology, and nuclear energy technology and sustainable energy, and determines multiple sub-topics that meet the topic selection needs of the user 310 as follows:

[0039] Sub-theme 1: Integration and complementary use of renewable energy technologies. Sub-theme description: Explore how to optimize the stability and efficiency of the energy system by mixing multiple renewable energy technologies (such as solar, wind, hydro, biomass, geothermal and tidal energy).

[0040] Sub-theme 2: Development and application of new energy storage technologies. Sub-theme description: Analyze how emerging energy storage solutions (including but not limited to battery technology, hydrogen energy storage, compressed air energy storage, etc.) promote the effective integration and utilization of renewable energy.

[0041] Sub-theme 3: Development and implementation of sustainable energy policies. Sub-theme description: Study how governments promote the development and adoption of renewable energy technologies through legislation, subsidies and other incentives.

[0042] Sub-theme 4: Community participation and localization of sustainable energy. Sub-theme description: Explore how the community level can promote energy transformation through participation and investment in renewable energy projects, and the applicability of this model in different regions around the world.

[0043] Sub-theme 5: Socio-economic impacts of renewable energy technologies. Sub-theme description: Analyze the impact of the deployment of renewable energy technologies on the job market, economic growth, and regional development.

[0044] Sub-theme 6: Energy Internet and Smart Energy Management. Sub-theme description: Explore how to manage and optimize distributed energy systems, including the production, storage and distribution of renewable energy, through advanced information technology and communication technology.

[0045] Finally, the topic selection platform 200 performs a multi-dimensional analysis on each sub-topic, and recommends the multiple sub-topics to the user 310 after ranking them according to the results of the analysis (S108). Specifically, the multiple dimensions for analyzing each sub-topic include, but are not limited to: research enthusiasm, cutting-edge nature, project investment, talent investment, paper output, patent output, national distribution, and institutional distribution. For each sub-topic, the above eight dimensions are analyzed, and based on the analysis results of each sub-topic, multiple sub-topics that meet the topic selection needs of the user 310 are ranked and recommended to the user 310. By performing a multi-dimensional analysis on each sub-topic, different dimensional information of each sub-topic can be obtained, which helps to gain an in-depth understanding of the current status, development trends, and potential opportunities of each sub-topic, so as to accurately rank multiple sub-topics and recommend them to the target object.

[0046] As mentioned in the background technology, although some existing topic selection tools can help researchers identify hot research areas and predict future research trends by analyzing large-scale scientific literature databases, these topic selection tools are still at the basic stage of data search and organization, and fail to deeply understand the complexity and diversity of research topic selection. In addition, they fail to fully consider the unique background and interests of each researcher, lack a personalized recommendation mechanism, and result in the recommended topics not matching the researcher's topic selection needs.

[0047] In view of this, according to this embodiment, when a user has a topic selection demand, the corresponding keyword can be input into the terminal device, and the topic selection platform 200 can obtain the keyword input by the user from the terminal device. Then, the topic selection platform 200 determines the technical field of the user's concern based on the user's historical reading data, laying the foundation for the subsequent personalized and accurate recommendation of sub-topics. Secondly, the topic selection platform 200 comprehensively considers the technical field of the user's concern and the input keyword, and determines multiple sub-topics that meet the user's topic selection needs, thereby realizing personalized sub-topic customization services, and providing data support for subsequent accurate topic recommendation. Finally, the topic selection platform 200 analyzes each sub-topic in multiple dimensions, and recommends multiple sub-topics to the user after accurately sorting them according to the results of the analysis, thereby achieving the purpose of personalized and accurate recommendation of sub-topics for users, and achieving the technical effect of being able to ensure that the sub-topics recommended by the topic selection platform are highly matched with the user's topic selection needs. This solves the technical problem that the existing topic recommendation method in the prior art cannot match the researcher's topic selection needs due to the lack of a personalized recommendation mechanism.

[0048] Optionally, the technical fields that the target object is interested in are determined based on the historical reading data of the target object, including: collecting the historical reading data of the target object; analyzing the historical reading data to determine the document collection that the target object has read, and counting the reading parameters of the target object for each document in the document collection; wherein the reading parameters include the number of readings and the reading time; for each document in the document collection, extracting the annotation content of the target object; determining the weight of each document in the document collection based on the reading parameters and the annotation content; wherein the weight is used to reflect the degree of attention paid by the target object to each document in the document collection; performing feature extraction on all documents in the document collection to determine the feature vector of each document; and determining the technical fields that the target object is interested in based on the feature vectors of all documents in the document collection and the corresponding weights.

[0049] Specifically, refer to Figure 2 As shown, the topic selection platform 200 provides functions such as document query, knowledge push, and communication community, and in each function, users are supported to view and annotate materials online. Therefore, the topic selection platform 200 can collect the historical reading data of the user 310 (corresponding to the target object) based on the viewing behavior of the user 310. Then, the historical reading data is analyzed to determine all the documents that the user 310 has read, forming a document collection, and then the reading parameters of each document in the document collection by the user 310 are counted. The reading parameters include the number of readings and the reading time.

[0050] In this embodiment, the document set read by user 310 includes document 1 to document m. For the jth document, the number of times user 310 reads it is recorded as n, and the reading time is recorded as t. j is any value from 1 to m. By counting the reading parameters of user 310 for each document in the document set, data support is provided for the subsequent user attention analysis of each document.

[0051] Furthermore, since the topic selection platform 200 supports the user 310 to annotate each document online, the user 310 can add annotations such as key marks, notes or comments online while reading the document. Such information reflects the user's interest in or degree of importance of the document content. Thus, the topic selection platform 200 can extract the annotation content of the user 310 for each document in the document collection. Afterwards, the topic selection platform 200 determines the weight ω of each document in the document collection based on the reading parameters and the annotation content. j . Weight ω j It is used to reflect the attention degree of user 310 to each document in the document collection, and the weight ω j The larger the value of , the higher the degree of attention paid by the user 310 to the document. In this way, the topic selection platform 200 not only takes into account the content of the document itself, but also incorporates the annotation information from the user's perspective.

[0052] Then, the topic selection platform 200 extracts features from all the documents in the document collection and determines the feature vector of each document. Specifically, for the jth document in the document collection, natural language processing (NLP) can be used to extract features from it and determine the feature vector v of the jth document. j Finally, the topic selection platform 200 selects the feature vectors v of all the documents in the document collection. j and the corresponding weight ω j , determine the technical field that the user 310 is interested in.

[0053] Therefore, the topic selection platform 200 can more accurately locate the technical fields that the user 310 is interested in by extracting features from the documents that the user 310 has read and combining the degree of interest that the user 310 has shown in these documents, thereby laying the foundation for subsequent personalized and accurate recommendations of sub-topics.

[0054] Optionally, the weight of each document in the document collection is determined according to the reading parameters and the annotated content, including: determining, according to the reading parameters, the total number of times N that the target object read the jth document in the document collection in the past history, the number of times n that the target object read the jth document in the document collection within a preset time period, the total reading time T for the jth document in the document collection in the past history, and the reading time t for the jth document in the document collection within the preset time period; and calculating the score of each document in the document collection according to the total number of readings N, the number of readings n, the total reading time T, the reading time t, and the annotated content by the following formula:

[0055]

[0056] In the formula, S j is the integral of the jth document in the document set; k1 and k2 are pre-set non-zero coefficients; k3 takes a non-zero coefficient value when the target object has annotated the jth document, and k3 takes a value of zero when the target object has not annotated the jth document;

[0057] The integral of each document in the document collection is mapped to a value in the interval [0,1], and the mapped value is determined as the weight of the corresponding document to obtain the weight of each document in the document collection; wherein the sum of the weights of all documents in the document collection is 1.

[0058] Specifically, the topic selection platform 200 can determine that the total number of times the user 310 has read the jth document in the document collection in the past history is N, and the total reading time is T, based on the statistical reading parameters. The parameter N can measure the long-term access frequency of the user 310 to the jth document. The parameter T can measure the total reading depth of the user 310 for the jth document, that is, the total amount of time spent on reading.

[0059] The topic selection platform 200 can also determine that the number of times the user 310 reads the jth document in the document collection within a preset time period is n, and the total reading time is t according to the statistical reading parameters. The parameter n can measure the frequency of the user 310's access to the jth document in a short period of time. The parameter t can measure the user 310's short-term reading depth of the jth document, that is, the time spent on reading in a short period of time. In this embodiment, the preset time period is, for example but not limited to, the most recent year.

[0060] The administrator of the topic selection platform 200 can configure the preset time period according to the actual application scenario. By introducing the concept of the preset time period, the topic selection platform 200 can capture and respond to the dynamic changes of user interests, which increases flexibility and real-time performance.

[0061] In addition, the topic selection platform 200 can also count whether the user 310 has marked the jth document.

[0062] Then, the topic selection platform 200 calculates the score of each document in the document collection according to the total reading times N, the reading times n, the total reading time T, the reading time t and the annotated content by the following formula:

[0063]

[0064] In the formula, S j is the integral of the jth document in the document collection; k1 and k2 are pre-set non-zero coefficients; k3 takes a non-zero coefficient value when the target object has annotated the jth document, and the value of k3 is zero when the target object has not annotated the jth document.

[0065] Furthermore, for each document’s score S j , map it to a value in the interval [0,1] through a corresponding function (such as the softmax function), and determine the mapped value as the weight of the corresponding document to obtain the weight ω of each document in the document collection j In this way, the relative importance or attractiveness of each document relative to the entire document set can be intuitively represented. Moreover, by ensuring that the sum of all document weights is equal to 1, the rationality of weight distribution is guaranteed.

[0066] Therefore, by comprehensively considering multiple reading behavior parameters, it is possible to accurately quantify and understand the user's interest and attention to each document.

[0067] Optionally, feature extraction is performed on all documents in the document collection to determine the feature vector of each document, including: extracting the subject, abstract and annotation content of the target object of each document in the document collection, and fusing the extracted information to form a feature text for each document; inputting the feature text into a pre-trained feature extraction model to output the feature vector of each document.

[0068] Specifically, for each document in the document collection, the topic selection platform 200 will extract the key parts of each document, including but not limited to the subject, abstract and annotation content of the document by the user 310. Afterwards, the extracted key information is integrated to form the feature text of each document. Thus, important feature information can be extracted for each document in the document collection, ensuring that the subsequent processing stage can focus on the most relevant and valuable content.

[0069] Then, the topic selection platform 200 inputs the feature text into a pre-trained feature extraction model and outputs a feature vector for each document. Specifically, the pre-trained feature extraction model is, for example, but not limited to, a convolutional neural network (CNN), a recursive neural network (RNN), or a Transformer architecture, which are trained with a large amount of data and can effectively extract meaningful information from the feature text and convert it into a feature vector in numerical form. By pre-training the feature text of each document into a feature extraction model, the feature vector of each document can be accurately output, providing data support for the subsequent positioning of the technical fields that the user 310 is concerned about.

[0070] Optionally, the technical field that the target object is concerned about is determined based on the feature vectors and corresponding weights of all documents in the document collection, including: determining a comprehensive feature vector based on the feature vectors and corresponding weights of all documents in the document collection; inputting the comprehensive feature vector into a trained classifier and outputting a multidimensional vector; wherein each element of the multidimensional vector corresponds to a technical field, and the numerical value corresponding to the element reflects the degree of concern of the target object to the technical field; and determining the technical field that the target object is concerned about based on the multidimensional vector.

[0071] Specifically, the topic selection platform 200 first determines a comprehensive feature vector based on the feature vectors and corresponding weights of all documents in the document collection. Figure 4 As shown, according to the feature vector v of all documents j and the corresponding weight ω j , through weighted summation, the feature vectors of all documents are combined into a comprehensive feature vector V j , the comprehensive feature vector V j It covers the user 310's interest in various technical fields involved in the entire document collection.

[0072] Then, the topic selection platform 200 converts the comprehensive feature vector V j As a pre-trained classifier, a multi-dimensional vector is output. In the multi-dimensional vector, each dimension (or element) corresponds to a specific technical field, and the value of each dimension directly reflects the degree of attention paid by the user 310 to the technical field. Finally, according to the multi-dimensional vector output by the classifier, the degree of attention paid by the user 310 to each technical field can be directly read out, and the preset number of technical fields with the highest ranking can be determined as the technical fields paid attention to by the user 310.

[0073] Therefore, the topic selection platform 200 can accurately identify the technical fields that the user 310 is interested in by analyzing the user 310's reading behavior and preferences for literature, laying the foundation for subsequent personalized and precise recommendation of sub-topics.

[0074] Optionally, based on the technical fields and subject terms that the target object is concerned with, multiple sub-topics that meet the target object's topic selection needs are determined, including: determining a keyword set from a pre-constructed vocabulary based on the technical fields and subject terms that the target object is concerned with; wherein there is similarity and correlation between each keyword in the keyword set and the subject terms; determining a first total frequency of each keyword in the keyword set appearing in all documents in the document library and a second total frequency of each keyword appearing in documents published in the document library within a preset time period; based on the first total frequency, determining multiple hot keywords from the keyword set; based on the second total frequency, determining multiple cutting-edge keywords from the keyword set; classifying the parts of speech of the multiple hot keywords and the multiple cutting-edge keywords, and combining the hot keywords and / or cutting-edge keywords with different parts of speech types based on the association between the keywords in the vocabulary to obtain multiple sub-topics that meet the target object's topic selection needs.

[0075] Specifically, the subject word input by user 310 described above is "sustainable energy", and the technical fields concerned by user 310 include solar energy technology, wind energy technology, hydropower technology, biomass energy technology, geothermal energy technology, tidal energy technology and nuclear energy technology. The topic selection platform 200 first matches "sustainable energy", "solar energy technology", "wind energy technology", "hydropower technology", "biomass energy technology", "geothermal energy technology", "tidal energy technology" and "nuclear energy technology" with each word in the pre-constructed vocabulary to determine multiple candidate keywords. Among them, the pre-constructed vocabulary is a database containing a large number of words and their related attributes (such as word frequency, semantic relationship, synonyms, etc.). The topic selection platform 200 searches for words matching "sustainable energy", "solar energy technology", "wind energy technology", "hydropower technology", "biomass energy technology", "geothermal energy technology", "tidal energy technology" and "nuclear energy technology" in the vocabulary by feature matching, thereby screening out words related to the subject word and the technical fields concerned by the user as candidate keywords.

[0076] Then, the topic selection platform 200 performs similarity expansion and relevance expansion on the selected candidate keywords, and determines a keyword set based on the expansion results. Specifically, for each candidate keyword, the candidate keyword can be similarly expanded by synonym search, near-synonym search, and hyponym search, so as to determine similar words that are close in meaning or related to the candidate keyword. For each candidate keyword, the co-occurrence frequency of words in a large amount of text can be analyzed by co-occurrence analysis, so as to determine the associated words that are associated with the candidate keyword. Afterwards, all candidate keywords, similar words and associated words of each candidate keyword are integrated to form a keyword set.

[0077] Further, for each keyword in the keyword set, the topic selection platform 200 determines the first total frequency of its appearance in all documents in the document library and the second total frequency of its appearance in documents published in the document library within a preset time period. Specifically, for each i keyword W in the keyword set, i , first determine the weighted frequency F of each document in the literature library by the following formula i,j :

[0078] F i,j =n i,j *ω j ;

[0079] Where n i,j Keyword W i In the literature L j The number of occurrences in ω j For the literature L j The weight of .

[0080] Then, the topic selection platform 200 further determines each keyword W i The total frequency of the first occurrence in all documents Σ j F i,j And each keyword W i The second total frequency Σ in the literature published in the preset time period 预设时间段 F i,j The preset time period may be flexibly configured according to actual needs, for example but not limited to the past two years.

[0081] Then, the topic selection platform 200 determines a plurality of hot keywords from the keyword set according to the first total frequency. And the topic selection platform 200 determines a plurality of cutting-edge keywords from the keyword set according to the second total frequency. Specifically, a plurality of hot keywords are, for example, renewable energy, energy system, hybrid energy, effective integration, community participation, policy making, energy transformation, job market, economic growth and energy Internet. These hot keywords represent concepts and entities that are currently receiving widespread attention in the field of sustainable energy. A plurality of cutting-edge keywords are, for example, integrated technology, complementary utilization, new energy storage, incentives, localization, regional development and smart energy management. These cutting-edge keywords reflect the latest progress and technical directions in the field of sustainable energy.

[0082] Then, the topic selection platform 200 classifies the parts of speech of multiple hot keywords and multiple frontier keywords, and determines the parts of speech of each hot keyword and frontier keyword. Among them, the parts of speech of keywords include nouns, verbs, adjectives, frontier hot keywords and their corresponding parts of speech. The following table 1 shows each hot keyword and its corresponding part of speech, and the following table 2 shows each frontier keyword and its corresponding part of speech.

[0083] Table 1

[0084] Hot Keywords Part of Speech Renewable Energy noun Energy Systems noun Hybrid Energy noun Effective integration verb Community Engagement verb Policy Development verb Energy Transition verb Job Market noun Economic growth noun Energy Internet noun

[0085] Table 2

[0086] Frontier Keywords Part of Speech Integrated Technology noun Complementary use verb New energy storage noun Incentives noun Localization adjective Regional Development noun Smart Energy Management noun

[0087] Then, the topic selection platform 200 combines hot keywords and / or cutting-edge keywords with different parts of speech types according to the association between the keywords in the vocabulary to obtain multiple sub-topics that meet the topic selection needs of the target object. Specifically, the construction process of each sub-topic is as follows:

[0088] (1) Sub-theme 1: Integration and complementary utilization of renewable energy technologies.

[0089] According to the correlation between the keywords, the hot keywords selected from Table 1 are: renewable energy, energy system and hybrid energy;

[0090] According to the correlation between the keywords, the frontier keywords selected from Table 2 are: integrated technology, complementary utilization;

[0091] According to the parts of speech of the selected hot keywords and cutting-edge keywords, “renewable energy”, “energy system” and “hybrid energy” with the part of speech of nouns are combined with “utilization” with the part of speech of verbs to obtain sub-theme 1 “integration and complementary utilization of renewable energy technologies”.

[0092] (2) Sub-theme 2: Development and application of new energy storage technologies.

[0093] According to the correlation between the keywords, the hot keywords selected from Table 1 are: renewable energy, effective integration;

[0094] According to the correlation between the keywords, the frontier keywords selected from Table 2 are: new energy storage;

[0095] According to the parts of speech of the selected hot keywords and cutting-edge keywords, “new energy storage” and “renewable energy” which are nouns are combined with “effective integration” which is a verb, and sub-topic 2 “Development and application of new energy storage technology” is obtained.

[0096] (3) Sub-theme 3: Formulation and implementation of sustainable energy policies.

[0097] According to the correlation between the keywords, the hot keywords selected from Table 1 are: sustainable energy, policy making;

[0098] According to the correlation between the keywords, the frontier keywords selected from Table 2 are: incentives;

[0099] According to the parts of speech of the selected hot keywords and cutting-edge keywords, “sustainable energy” and “incentives” which are nouns are combined with “policy making” which is a verb, and sub-theme 3 “Formulation and implementation of sustainable energy policies” is obtained.

[0100] (4) Sub-theme 4: Community participation and localization of sustainable energy.

[0101] According to the correlation between the keywords, the hot keywords selected from Table 1 are: community participation, sustainable energy, and energy transformation;

[0102] According to the correlation between the keywords, the frontier keywords selected from Table 2 are: localization;

[0103] According to the parts of speech of the selected hot keywords and cutting-edge keywords, “sustainable energy” which is a noun is combined with “community participation” and “energy transformation” which are verbs, and “localization” which is an adjective, to obtain sub-theme 4 “community participation and localization of sustainable energy”.

[0104] (5) Sub-theme 5: Socio-economic impacts of renewable energy technologies.

[0105] According to the correlation between the keywords, the hot keywords selected from Table 1 are: renewable energy technology, job market, economic growth;

[0106] According to the correlation between the keywords, the frontier keywords selected from Table 2 are: regional development;

[0107] According to the parts of speech of the selected hot keywords and cutting-edge keywords, “renewable energy technology”, “employment market”, “economic growth” and “regional development”, all of which are nouns, are combined to obtain sub-theme 5 “socio-economic impact of renewable energy technology”.

[0108] (6) Sub-theme 6: Energy Internet and Smart Energy Management.

[0109] According to the correlation between the keywords, the hot keywords selected from Table 1 are: energy internet;

[0110] According to the correlation between the keywords, the frontier keywords selected from Table 2 are: smart energy management;

[0111] According to the parts of speech of the selected hot keywords and cutting-edge keywords, “Energy Internet” and “Smart Energy Management”, both of which are nouns, are combined to obtain sub-topic 6 “Energy Internet and Smart Energy Management”.

[0112] Thus, through the above analysis, by carefully selecting hot keywords and cutting-edge keywords, and combining their word characteristics and internal associations, a sub-topic with both depth and breadth is formed. This combination not only reflects the current research hotspots, but also points to the trend of future development, which helps to deeply understand and promote the development of renewable energy and related technologies, and ensures that each sub-topic determined is highly matched with the topic selection needs of users 310.

[0113] Optionally, each sub-topic is analyzed in multiple dimensions, and the multiple sub-topics are ranked according to the analysis results and recommended to the target object, including: analyzing each sub-topic in multiple dimensions, determining the analysis results of each sub-topic at each time node in different dimensions according to the analysis results; determining the feature matrix of each sub-topic according to the analysis results of each sub-topic at each time node in different dimensions Among them, C is the number of analysis dimensions of each sub-topic; T is the number of time nodes of each sub-topic; according to the feature matrix of each sub-topic Determine the selection probability value of each sub-topic; and rank multiple sub-topics according to the selection probability values ​​and recommend them to the target object.

[0114] Specifically, for each sub-topic, analysis is performed from eight dimensions, for example but not limited to, including research heat analysis, cutting-edge, project investment, talent investment, paper output, patent output, national distribution, and institutional distribution. For each sub-topic, the analysis results of each time node are determined under each dimension. Among them, the time range of the analysis is several years, and the unit is year, then each time node is a different year. For example, the time range of the analysis is from 2013 to 2023, and the time nodes within this time range are 2013, 2014, 2015, 2016, 2017, 2018, 2019, 2020, 2021, 2022, and 2023.

[0115] In this embodiment, sub-topic 1 is taken as an example to illustrate how to determine the analysis results of each time node in different dimensions. The principles of other sub-topics are the same and will not be repeated here.

[0116] For sub-topic 1, first determine the analysis results of sub-topic 1 at each time node under the dimension of "research heat analysis", recorded as (a 11 , a 12 , a 13 , a 14 , a 15 , a 16 , a 17 , a 18 , a 19 , a 110), where a 11 ~a 110 These are the analysis results of sub-topic 1 under the dimension of "Research Heat Analysis" in 2013, 2014, 2015, 2016, 2017, 2018, 2019, 2020, 2021, 2022 and 2023.

[0117] In this example, the analysis results of sub-topic 1 in 2013 under the dimension of "research heat analysis" are expanded to explain a 11 The process of determining 12 ~a 110 The determination steps are similar to a 11 The same, no further description is given here. 11 is the annual research intensity of sub-topic 1, a 11 It is equal to the sum of the total frequencies of all hot keywords (i.e., renewable energy, energy system, and hybrid energy) and all cutting-edge keywords (i.e., integrated technology and complementary utilization) included in subtopic 1 in all documents in the literature library.

[0118] Among them, the research popularity analysis mainly presents the changing trends of the word frequencies of each sub-topic by year. In one graph, the changing trends of the word frequencies of all sub-topics are presented, so that users can intuitively analyze the popularity evolution of the sub-topic.

[0119] Then, the analysis results of sub-topic 1 at each time node under the dimension of “frontier” are determined, recorded as (a 21 , a 22 , a 23 , a 24 , a 25 , a 26 , a 27 , a 28 , a 29 , a 210 ), where a 21 ~a 210 These are the analysis results of sub-theme 1 under the "frontier" dimension in 2013, 2014, 2015, 2016, 2017, 2018, 2019, 2020, 2021, 2022 and 2023.

[0120] In this example, the analysis results of sub-topic 1 in the dimension of “frontier” in 2013 are described in detail. 21 The process of determining 22 ~a 210 The determination steps are similar to a 21 The same, no further description is given here. 21 is the growth rate of sub-topic 1, a21 =The number of all hot keywords and all cutting-edge keywords included in sub-topic 1 that appeared in the literature published in 2013 / the total number of all hot keywords and all cutting-edge keywords included in sub-topic 1 that appeared in the literature published in 2013 and before.

[0121] Among them, the cutting-edge analysis mainly presents the changing trend of the growth rate of each sub-topic keyword on an annual basis to check whether there are sudden changes in the keywords within a certain period of time, so as to identify new research hotspots and directions and conduct cutting-edge analysis.

[0122] Then, determine the analysis results of each time node of sub-topic 1 under the dimension of "project investment", recorded as (a 31 , a 32 , a 33 , a 34 , a 35 , a 36 , a 37 , a 38 , a 39 , a 310 ), where a 31 ~a 310 These are the analysis results of sub-theme 1 under the dimension of "project investment" in 2013, 2014, 2015, 2016, 2017, 2018, 2019, 2020, 2021, 2022 and 2023.

[0123] In this example, the analysis results of sub-topic 1 in the dimension of "project investment" in 2013 are described in detail. 31 The process of determining 32 ~a 310 The determination steps are similar to a 31 The same, no further description is given here. 31 is the annual project input of sub-theme 1, a 31 =The total number of new projects added to Sub-Theme 1 in 2013.

[0124] Among them, the project investment analysis mainly presents the changing trend of the number of projects in each sub-theme on an annual basis, in order to analyze whether each sub-theme has been invested in a key area.

[0125] Then, determine the analysis results of sub-topic 1 at each time node under the dimension of “talent investment”, recorded as (a 41 , a 42 , a 43 , a 44 , a 45 , a 46 , a 47 , a 48 , a49 , a 410 ), where a 41 ~a 410 They are the analysis results of sub-theme 1 under the dimension of “talent investment” in 2013, 2014, 2015, 2016, 2017, 2018, 2019, 2020, 2021, 2022 and 2023.

[0126] In this example, the analysis results of sub-topic 1 in the dimension of “talent investment” in 2013 are described in detail. 41 The process of determining 42 ~a 410 The determination steps are similar to a 41 The same, no further description is given here. 41 is the annual talent input of sub-theme 1, a 41 =Total number of talents invested in sub-theme 1 in 2013.

[0127] Among them, the talent investment analysis mainly presents the changing trend of the number of talent investments in each sub-theme on an annual basis, in order to analyze whether each sub-theme has been invested in a key area.

[0128] Then, the analysis results of sub-topic 1 at each time node under the dimension of “paper output” are determined, recorded as (a 51 , a 52 , a 53 , a 54 , a 55 , a 56 , a 57 , a 58 , a 59 , a 510 ), where a 51 ~a 510 These are the analysis results of sub-topic 1 under the "paper output" dimension in 2013, 2014, 2015, 2016, 2017, 2018, 2019, 2020, 2021, 2022 and 2023.

[0129] In this example, the analysis results of sub-topic 1 in the dimension of "paper output" in 2013 are described in detail. 51 The process of determining 52 ~a 510 The determination steps are similar to a 51 The same, no further description is given here. 51 is the annual output of papers for sub-topic 1, a 51 =The total number of papers produced by sub-topic 1 in 2013.

[0130] Among them, the paper output analysis mainly presents the changing trend of the number of papers input in each sub-topic on an annual basis, in order to analyze whether each sub-topic is a research focus.

[0131] Then, the analysis results of sub-topic 1 at each time node under the dimension of “patent output” are determined, recorded as (a 61 , a 62 , a 63 , a 64 , a 65 , a 66 , a 67 , a 68 , a 69 , a 610 ), where a 61 ~a 610 These are the analysis results of sub-topic 1 under the "patent output" dimension in 2013, 2014, 2015, 2016, 2017, 2018, 2019, 2020, 2021, 2022 and 2023.

[0132] In this example, the analysis results of sub-topic 1 in the dimension of "patent output" in 2013 are described in detail. 61 The process of determining 62 ~a 610 The determination steps are similar to a 61 The same, no further description is given here. 61 is the annual patent output of sub-topic 1, a 61 =The total number of patents produced by sub-topic 1 in 2013.

[0133] Among them, the patent output analysis mainly presents the changing trend of the number of patent inputs in each sub-topic on an annual basis, in order to analyze whether each sub-topic is the focus of industrialization.

[0134] Then, the analysis results of sub-topic 1 at each time node under the dimension of “national distribution” are determined, recorded as (a 71 , a 72 , a 73 , a 74 , a 75 , a 76 , a 77 , a 78 , a 79 , a 710 ), where a 71 ~a 710These are the analysis results of sub-theme 1 under the “National Distribution” dimension in 2013, 2014, 2015, 2016, 2017, 2018, 2019, 2020, 2021, 2022 and 2023.

[0135] In this example, the analysis results of sub-topic 1 in 2013 under the dimension of “country distribution” are described in detail. 71 The process of determining 72 ~a 710 The determination steps are similar to a 71 The same, no further description is given here. 71 is the annual national input and annual national output of sub-theme 1, a 71 =(national input, national output). Among them, national input refers to the total number of projects invested by the state in sub-topic 1 in 2013, and national output refers to the total number of papers and patents produced by sub-topic 1 in 2013.

[0136] Among them, the national input distribution analysis mainly presents the national input distribution of each sub-topic, so as to analyze which countries are focusing on each sub-topic. The national output distribution analysis mainly presents the national scientific research output distribution of each sub-topic, so as to analyze which countries are focusing on each sub-topic.

[0137] Then, the analysis results of sub-topic 1 at each time node under the dimension of “institutional distribution” are determined and recorded as (a 81 , a 82 , a 83 , a 84 , a 85 , a 86 , a 87 , a 88 , a 89 , a 810 ), where a 81 ~a 810 These are the analysis results of sub-theme 1 under the dimension of "institutional distribution" in 2013, 2014, 2015, 2016, 2017, 2018, 2019, 2020, 2021, 2022 and 2023.

[0138] In this example, the analysis results of sub-topic 1 in the dimension of “institutional distribution” in 2013 are described in detail. 81 The process of determining 82 ~a 810 The determination steps are similar to a 81 The same, no further description is given here. 81 is the annual institutional input and annual institutional output of sub-theme 1, a81 =(institutional input, institutional output). Institutional input refers to the total number of projects invested by the institution in sub-topic 1 in 2013, and institutional output refers to the total number of papers and patents produced by sub-topic 1 in 2013.

[0139] Among them, the institutional input distribution analysis mainly presents the distribution of institutional inputs in each sub-topic, in order to analyze which institutions are focusing on each sub-topic. The institutional output distribution analysis mainly presents the distribution of scientific research outputs in each sub-topic, in order to analyze which institutions are focusing on each sub-topic.

[0140] Furthermore, according to the analysis results of each sub-topic at each time node in different dimensions, the feature matrix of each sub-topic is determined.

[0141]

[0142] Wherein, C is the number of analysis dimensions of each sub-topic, and in this embodiment, C=8; T is the number of time nodes of each sub-topic, and in this embodiment, T=10.

[0143] Then, according to the feature matrix of each sub-topic Determine the selection probability value of each sub-topic. Specifically, the selection probability value of each sub-topic can be calculated by a machine learning algorithm to ensure that information of different dimensions and time points can be fairly reflected. Finally, sort them according to the calculated selection probability value of each sub-topic. The result of the sorting directly determines the order of recommendation to user 310. Among them, sub-topics with high selection probability values ​​are generally more in line with the needs of the target object or have better expected effects, and will be recommended first.

[0144] Therefore, not only can the sub-topic recommendation process be made more objective and accurate, but also the quality and acceptance of the recommendation results can be improved.

[0145] Optionally, based on the feature matrix for each sub-topic Determine the selection probability value of each sub-topic, including: determine the feature matrix of each sub-topic The covariance matrix P of is:

[0146]

[0147] According to the pre-built signal projection matrix and covariance matrix P, extract the features x of each subtopic j :

[0148]

[0149] Among them, x jrepresents the feature of the j-th subtopic, and M represents the feature x j The logarithm of the characteristic, and diag(A) means returning the diagonal elements of the matrix A and forming a vector;

[0150] Use the jth binary classifier to classify the feature x of the jth sub-topic j Perform binary classification and determine the probability value of selecting the jth sub-topic.

[0151] Specifically, first determine the feature matrix of each sub-topic The covariance matrix P of is:

[0152]

[0153] Then, according to the pre-built signal projection matrix and covariance matrix P, extract the features x of each subtopic j :

[0154]

[0155] Among them, x j represents the feature of the j-th subtopic, and M represents the feature x j The logarithm of the characteristic of , and diag(A) means returning the diagonal elements of the matrix A and forming a vector.

[0156] Then, refer to Figure 5 As shown, for the j-th sub-topic, a binary classifier is used to classify the feature x of the j-th sub-topic j Perform binary classification to determine the probability of selecting the jth sub-topic. The two categories of each binary classifier are the selected category and the non-selected category.

[0157] In this embodiment, the binary classifiers corresponding to all sub-topics share a signal projection matrix W j , the signal projection matrix W j This can be determined by:

[0158] Step 1: The topic selection platform 200 determines a sample set of feature matrices based on historical recommendation records Where i = 1 or 2. is the feature matrix (n=1~N1) corresponding to a category of the jth binary classifier (i.e., binary classifier j); is the feature matrix (n=1~N2) corresponding to the other category of the j-th binary classifier (ie, binary classifier j).

[0159] Step 2: For each feature matrix Calculate the feature matrix The covariance matrix of is:

[0160]

[0161] Step 3: Calculate separately The mean and The mean As the covariance mean of matrix samples of two different categories of binary classifier j:

[0162]

[0163] Step 4: Construct a signal difference maximization model and calculate the signal projection matrix W j :

[0164]

[0165] Therefore, through the above operations, the signal projection matrix W can be determined j , to extract the features x of each sub-topic j Provide a data basis. In addition, according to this embodiment, a storage medium is also provided. The storage medium includes a stored program, wherein when the program is running, the processor executes any one of the above methods.

[0166] In this application, when a user has a topic selection requirement, the corresponding keyword can be input into the terminal device, and the topic selection platform can obtain the keyword input by the user from the terminal device. Then, the topic selection platform determines the technical field of the user's concern based on the user's historical reading data, laying the foundation for the subsequent personalized and accurate recommendation of sub-topics. Secondly, the topic selection platform comprehensively considers the technical field of the user's concern and the input keyword, and determines multiple sub-topics that meet the user's topic selection requirements, thereby realizing personalized sub-topic customization services and providing data support for subsequent accurate topic recommendations. Finally, the topic selection platform analyzes each sub-topic in multiple dimensions, and recommends multiple sub-topics to users after accurately sorting them according to the results of the analysis, thereby achieving the purpose of personalized and accurate recommendation of sub-topics for users, and achieving the technical effect of being able to ensure that the sub-topics recommended by the topic selection platform are highly matched with the user's topic selection requirements. This solves the technical problem that the existing topic recommendation method in the prior art cannot match the researcher's topic selection requirements due to the lack of a personalized recommendation mechanism.

[0167] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described action sequence, because according to the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0168] Example 2

[0169] Figure 6 The structure diagram of the topic selection device 600 based on artificial intelligence according to the present embodiment is shown. The topic selection device 600 includes: an acquisition module 610, which is used to acquire the subject words input by the target object; a first determination module 620, which is used to determine the technical field that the target object is concerned about based on the historical reading data of the target object; a second determination module 630, which is used to determine multiple sub-topics that meet the topic selection needs of the target object based on the technical field and subject words that the target object is concerned about; an analysis and recommendation module 640, which is used to analyze each sub-topic in multiple dimensions, and recommend multiple sub-topics to the target object after sorting them according to the analysis results.

[0170] Optionally, the first determination module 620 is specifically used to: collect historical reading data of the target object; analyze the historical reading data to determine the document set that the target object has read, and count the reading parameters of the target object for each document in the document set; wherein the reading parameters include the number of readings and the reading time; for each document in the document set, extract the annotation content of the target object; determine the weight of each document in the document set based on the reading parameters and the annotation content; wherein the weight is used to reflect the degree of attention paid by the target object to each document in the document set; perform feature extraction on all documents in the document set to determine the feature vector of each document; and determine the technical field of interest of the target object based on the feature vectors of all documents in the document set and the corresponding weights.

[0171] Optionally, the weight of each document in the document collection is determined according to the reading parameters and the annotated content, including: determining, according to the reading parameters, the total number of times N that the target object read the jth document in the document collection in the past history, the number of times n that the target object read the jth document in the document collection within a preset time period, the total reading time T for the jth document in the document collection in the past history, and the reading time t for the jth document in the document collection within the preset time period; and calculating the score of each document in the document collection according to the total number of readings N, the number of readings n, the total reading time T, the reading time t, and the annotated content by the following formula:

[0172]

[0173] In the formula, S j is the integral of the jth document in the document set; k1 and k2 are pre-set non-zero coefficients; k3 takes a non-zero coefficient value when the target object has annotated the jth document, and k3 takes a value of zero when the target object has not annotated the jth document;

[0174] The integral of each document in the document collection is mapped to a value in the interval [0,1], and the mapped value is determined as the weight of the corresponding document to obtain the weight of each document in the document collection; wherein the sum of the weights of all documents in the document collection is 1.

[0175] Optionally, the technical field that the target object is concerned about is determined based on the feature vectors and corresponding weights of all documents in the document collection, including: determining a comprehensive feature vector based on the feature vectors and corresponding weights of all documents in the document collection; inputting the comprehensive feature vector into a trained classifier and outputting a multidimensional vector; wherein each element of the multidimensional vector corresponds to a technical field, and the numerical value corresponding to the element reflects the degree of concern of the target object to the technical field; and determining the technical field that the target object is concerned about based on the multidimensional vector.

[0176] Optionally, the second determination module 630 is specifically used to: determine a keyword set from a pre-constructed vocabulary based on the technical fields and subject terms that the target object is concerned about; wherein there are similarities and correlations between each keyword in the keyword set and the subject terms; determine a first total frequency of each keyword in the keyword set appearing in all documents in the document library and a second total frequency of each keyword appearing in documents published in the document library within a preset time period; determine a plurality of hot keywords from the keyword set based on the first total frequency; determine a plurality of cutting-edge keywords from the keyword set based on the second total frequency; classify the parts of speech of the plurality of hot keywords and the plurality of cutting-edge keywords, and combine the hot keywords and / or cutting-edge keywords with different parts of speech types based on the association between the keywords in the vocabulary to obtain a plurality of sub-topics that meet the topic selection needs of the target object.

[0177] Optionally, the analysis and recommendation module 640 is specifically used to: analyze each sub-topic in multiple dimensions, and determine the analysis results of each sub-topic at each time node in different dimensions according to the analysis results; determine the feature matrix of each sub-topic according to the analysis results of each sub-topic at each time node in different dimensions. Among them, C is the number of analysis dimensions of each sub-topic; T is the number of time nodes of each sub-topic; according to the feature matrix of each sub-topic Determine the selection probability value of each sub-topic; and rank multiple sub-topics according to the selection probability values ​​and recommend them to the target object.

[0178] Optionally, based on the feature matrix for each sub-topic Determine the selection probability value of each sub-topic, including: determine the feature matrix of each sub-topic The covariance matrix P of is:

[0179]

[0180] According to the pre-built signal projection matrix and covariance matrix P, extract the features x of each subtopic j :

[0181]

[0182] Among them, x j represents the feature of the j-th subtopic, and M represents the feature x j The logarithm of the characteristic, and diag(A) means returning the diagonal elements of the matrix A and forming a vector;

[0183] Use the jth binary classifier to classify the feature x of the jth sub-topic j Perform binary classification and determine the probability value of selecting the jth sub-topic.

[0184] Thus, according to this embodiment, when a user has a topic selection requirement, the corresponding keyword can be input into the terminal device, and the topic selection platform can obtain the keyword input by the user from the terminal device. Then, the topic selection platform determines the technical field of the user's concern based on the user's historical reading data, laying the foundation for the subsequent personalized and accurate recommendation of sub-topics. Secondly, the topic selection platform comprehensively considers the technical field of the user's concern and the input keyword, and determines multiple sub-topics that meet the user's topic selection requirements, thereby realizing personalized sub-topic customization services, and providing data support for subsequent accurate topic recommendations. Finally, the topic selection platform analyzes each sub-topic in multiple dimensions, and recommends multiple sub-topics to the user after accurately sorting them according to the results of the analysis, thereby achieving the purpose of personalized and accurate recommendation of sub-topics for users, and achieving the technical effect of being able to ensure that the sub-topics recommended by the topic selection platform are highly matched with the user's topic selection requirements. Thus, the technical problem that the existing topic recommendation method in the prior art cannot match the researcher's topic selection requirements due to the lack of a personalized recommendation mechanism is solved.

[0185] Example 3

[0186] Figure 7The artificial intelligence-based topic selection device 700 according to the present embodiment is shown, comprising: a processor 710; and a memory 720, connected to the processor 710, for providing instructions for the processor 710 to process the following processing steps: obtaining a keyword input by a target object; determining the technical field of interest to the target object based on the historical reading data of the target object; determining multiple sub-topics that meet the topic selection needs of the target object based on the technical fields and keywords of interest to the target object; performing a multi-dimensional analysis on each sub-topic, and ranking the multiple sub-topics based on the results of the analysis and recommending them to the target object.

[0187] Optionally, the technical fields that the target object is interested in are determined based on the historical reading data of the target object, including: collecting the historical reading data of the target object; analyzing the historical reading data to determine the document collection that the target object has read, and counting the reading parameters of the target object for each document in the document collection; wherein the reading parameters include the number of readings and the reading time; for each document in the document collection, extracting the annotation content of the target object; determining the weight of each document in the document collection based on the reading parameters and the annotation content; wherein the weight is used to reflect the degree of attention paid by the target object to each document in the document collection; performing feature extraction on all documents in the document collection to determine the feature vector of each document; and determining the technical fields that the target object is interested in based on the feature vectors of all documents in the document collection and the corresponding weights.

[0188] Optionally, the weight of each document in the document collection is determined according to the reading parameters and the annotated content, including: determining, according to the reading parameters, the total number of times N that the target object read the jth document in the document collection in the past history, the number of times n that the target object read the jth document in the document collection within a preset time period, the total reading time T for the jth document in the document collection in the past history, and the reading time t for the jth document in the document collection within the preset time period; and calculating the score of each document in the document collection according to the total number of readings N, the number of readings n, the total reading time T, the reading time t, and the annotated content by the following formula:

[0189]

[0190] In the formula, S j is the integral of the jth document in the document set; k1 and k2 are pre-set non-zero coefficients; k3 takes a non-zero coefficient value when the target object has annotated the jth document, and k3 takes a value of zero when the target object has not annotated the jth document;

[0191] The integral of each document in the document collection is mapped to a value in the interval [0,1], and the mapped value is determined as the weight of the corresponding document to obtain the weight of each document in the document collection; wherein the sum of the weights of all documents in the document collection is 1.

[0192] Optionally, the technical field that the target object is concerned about is determined based on the feature vectors and corresponding weights of all documents in the document collection, including: determining a comprehensive feature vector based on the feature vectors and corresponding weights of all documents in the document collection; inputting the comprehensive feature vector into a trained classifier and outputting a multidimensional vector; wherein each element of the multidimensional vector corresponds to a technical field, and the numerical value corresponding to the element reflects the degree of concern of the target object to the technical field; and determining the technical field that the target object is concerned about based on the multidimensional vector.

[0193] Optionally, based on the technical fields and subject terms that the target object is concerned with, multiple sub-topics that meet the target object's topic selection needs are determined, including: determining a keyword set from a pre-constructed vocabulary based on the technical fields and subject terms that the target object is concerned with; wherein there is similarity and correlation between each keyword in the keyword set and the subject terms; determining a first total frequency of each keyword in the keyword set appearing in all documents in the document library and a second total frequency of each keyword appearing in documents published in the document library within a preset time period; based on the first total frequency, determining multiple hot keywords from the keyword set; based on the second total frequency, determining multiple cutting-edge keywords from the keyword set; classifying the parts of speech of the multiple hot keywords and the multiple cutting-edge keywords, and combining the hot keywords and / or cutting-edge keywords with different parts of speech types based on the association between the keywords in the vocabulary to obtain multiple sub-topics that meet the target object's topic selection needs.

[0194] Optionally, each sub-topic is analyzed in multiple dimensions, and the multiple sub-topics are ranked according to the analysis results and recommended to the target object, including: analyzing each sub-topic in multiple dimensions, determining the analysis results of each sub-topic at each time node in different dimensions according to the analysis results; determining the feature matrix of each sub-topic according to the analysis results of each sub-topic at each time node in different dimensions Among them, C is the number of analysis dimensions of each sub-topic; T is the number of time nodes of each sub-topic; according to the feature matrix of each sub-topic Determine the selection probability value of each sub-topic; and rank multiple sub-topics according to the selection probability values ​​and recommend them to the target object.

[0195] Optionally, based on the feature matrix for each sub-topic Determine the selection probability value of each sub-topic, including: determine the feature matrix of each sub-topic The covariance matrix P of is:

[0196]

[0197] According to the pre-built signal projection matrix and covariance matrix P, extract the features x of each subtopic j :

[0198]

[0199] Among them, x j represents the feature of the j-th subtopic, and M represents the feature x j The logarithm of the characteristic, and diag(A) means returning the diagonal elements of the matrix A and forming a vector;

[0200] Use the jth binary classifier to classify the feature x of the jth sub-topic j Perform binary classification and determine the probability value of selecting the jth sub-topic.

[0201] Thus, according to this embodiment, when a user has a topic selection requirement, the corresponding keyword can be input into the terminal device, and the topic selection platform can obtain the keyword input by the user from the terminal device. Then, the topic selection platform determines the technical field of the user's concern based on the user's historical reading data, laying the foundation for the subsequent personalized and accurate recommendation of sub-topics. Secondly, the topic selection platform comprehensively considers the technical field of the user's concern and the input keyword, and determines multiple sub-topics that meet the user's topic selection requirements, thereby realizing personalized sub-topic customization services, and providing data support for subsequent accurate topic recommendations. Finally, the topic selection platform analyzes each sub-topic in multiple dimensions, and recommends multiple sub-topics to the user after accurately sorting them according to the results of the analysis, thereby achieving the purpose of personalized and accurate recommendation of sub-topics for users, and achieving the technical effect of being able to ensure that the sub-topics recommended by the topic selection platform are highly matched with the user's topic selection requirements. Thus, the technical problem that the existing topic recommendation method in the prior art cannot match the researcher's topic selection requirements due to the lack of a personalized recommendation mechanism is solved.

[0202] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0203] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0204] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0205] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0206] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0207] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, RandomAccess Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0208] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A topic selection method based on artificial intelligence, characterized in that: include: Get the subject words input by the target object; Determining the technical field of interest to the target object based on the target object's historical reading data; Determine a keyword set from a pre-built word library according to the technical field that the target object is interested in and the subject word; wherein each keyword in the keyword set has similarity and relevance with the subject word; Determine a first total frequency of each keyword in the keyword set appearing in all documents in a document library and a second total frequency of each keyword appearing in documents in the document library published within a preset time period; According to the first total frequency, a plurality of hot keywords are determined from the keyword set; according to the second total frequency, a plurality of cutting-edge keywords are determined from the keyword set; Classifying the parts of speech of the multiple hot keywords and the multiple cutting-edge keywords, and combining the hot keywords and / or cutting-edge keywords with different parts of speech types according to the association relationship between the keywords in the vocabulary, to obtain multiple sub-topics that meet the topic selection needs of the target object; Conduct multi-dimensional analysis on each sub-topic, and determine the analysis results of each sub-topic at each time node in different dimensions based on the analysis results; Determine the feature matrix of each sub-topic based on the analysis results of each time node in different dimensions Where C is the number of analysis dimensions of each sub-topic; T is the number of time nodes of each sub-topic; According to the characteristic matrix of each sub-topic Determine the selection probability value of each sub-topic; According to the selection probability values, the multiple sub-topics are sorted and recommended to the target object.

2. The method according to claim 1, characterized in that Determining the technical field that the target object is interested in based on the historical reading data of the target object includes: Collecting historical reading data of the target object; Analyze the historical reading data to determine the document set that the target object has read, and count the reading parameters of the target object for each document in the document set; wherein the reading parameters include the number of readings and the reading time; For each document in the document collection, extract the annotation content of the target object; Determine the weight of each document in the document collection according to the reading parameters and the annotated content; wherein the weight is used to reflect the degree of attention of the target object to each document in the document collection; Extracting features from all documents in the document collection to determine a feature vector for each document; The technical field that the target object is interested in is determined according to the feature vectors and corresponding weights of all the documents in the document collection.

3. The method according to claim 2, characterized in that Determining the weight of each document in the document collection according to the reading parameters and the annotated content includes: According to the reading parameters, determine the total number of times N that the target object has read the jth document in the document collection in the past history, the number of times n that the target object has read the jth document in the document collection within a preset time period, the total reading time T for the jth document in the document collection in the past history, and the reading time t for the jth document in the document collection within the preset time period; According to the total number of readings N, the number of readings n, the total reading time T, the reading time t and the annotated content, the score of each document in the document collection is calculated by the following formula: In the formula, S j is the integral of the jth document in the document set; k1 and k2 are pre-set non-zero coefficients; k3 takes a non-zero coefficient value when the target object has marked the jth document, and takes a value of zero when the target object has not marked the jth document; The integral of each document in the document collection is mapped to a numerical value in the interval [0,1], and the mapped numerical value is determined as the weight of the corresponding document, thereby obtaining the weight of each document in the document collection; wherein the sum of the weights of all documents in the document collection is 1.

4. The method according to claim 2, characterized in that: Determining the technical field of interest to the target object based on the feature vectors and corresponding weights of all documents in the document collection includes: Determine a comprehensive feature vector according to the feature vectors and corresponding weights of all documents in the document collection; Input the comprehensive feature vector into the trained classifier to output a multidimensional vector; wherein each element of the multidimensional vector corresponds to a technical field, and the value corresponding to the element reflects the degree of attention paid by the target object to the technical field; The technical field that the target object is interested in is determined according to the multi-dimensional vector.

5. The method according to claim 1, characterized in that The characteristic matrix of each sub-topic Determine the selection probability value for each sub-topic, including: Determine the feature matrix for each sub-topic The covariance matrix P of is: According to the pre-built signal projection matrix and the covariance matrix P, extract the features x of each sub-topic j : Among them, x j represents the feature of the j-th subtopic, and M represents the feature x j The logarithm of the characteristic, and diag(A) means returning the diagonal elements of the matrix A and forming a vector; Use the jth binary classifier to classify the feature x of the jth sub-topic j Perform binary classification and determine the probability value of selecting the jth sub-topic.

6. A storage medium, characterized in that: The storage medium includes a stored program, wherein when the program is run, the processor executes the method according to any one of claims 1 to 5.

7. A topic selection device based on artificial intelligence, comprising: An acquisition module is used to acquire the subject words input by the target object; A first determination module is used to determine the technical field that the target object is interested in based on the historical reading data of the target object; A second determination module is used to determine a keyword set from a pre-constructed vocabulary based on the technical field of interest to the target object and the subject words; wherein there are similarities and correlations between each keyword in the keyword set and the subject words; determine a first total frequency of each keyword in the keyword set appearing in all documents in the document library and a second total frequency of each keyword appearing in documents published in the document library within a preset time period; determine a plurality of hot keywords from the keyword set based on the first total frequency; determine a plurality of cutting-edge keywords from the keyword set based on the second total frequency; classify the parts of speech of the plurality of hot keywords and the plurality of cutting-edge keywords, and combine the hot keywords and / or cutting-edge keywords with different parts of speech types based on the correlation between the keywords in the vocabulary to obtain a plurality of sub-topics that meet the topic selection needs of the target object; The analysis and recommendation module is used to analyze each sub-topic in multiple dimensions, and determine the analysis results of each sub-topic at each time node in different dimensions based on the analysis results; according to the analysis results of each sub-topic at each time node in different dimensions, determine the feature matrix of each sub-topic Among them, C is the number of analysis dimensions of each sub-topic; T is the number of time nodes of each sub-topic; according to the feature matrix of each sub-topic Determine a selection probability value for each sub-topic; and, based on the selection probability values, sort the multiple sub-topics and recommend them to the target object.

8. A topic selection device based on artificial intelligence, characterized in that: include: processor; as well as A memory, connected to the processor, configured to provide the processor with instructions for processing the following processing steps: Get the subject words input by the target object; Determining the technical field of interest to the target object based on the target object's historical reading data; Determine a keyword set from a pre-built word library according to the technical field that the target object is interested in and the subject word; wherein each keyword in the keyword set has similarity and relevance with the subject word; Determine a first total frequency of each keyword in the keyword set appearing in all documents in a document library and a second total frequency of each keyword appearing in documents in the document library published within a preset time period; According to the first total frequency, a plurality of hot keywords are determined from the keyword set; according to the second total frequency, a plurality of cutting-edge keywords are determined from the keyword set; Classifying the parts of speech of the multiple hot keywords and the multiple cutting-edge keywords, and combining the hot keywords and / or cutting-edge keywords with different parts of speech types according to the association relationship between the keywords in the vocabulary, to obtain multiple sub-topics that meet the topic selection needs of the target object; Conduct multi-dimensional analysis on each sub-topic, and determine the analysis results of each sub-topic at each time node in different dimensions based on the analysis results; Determine the feature matrix of each sub-topic based on the analysis results of each time node in different dimensions Where C is the number of analysis dimensions of each sub-topic; T is the number of time nodes of each sub-topic; According to the characteristic matrix of each sub-topic Determine the selection probability value of each sub-topic; According to the selection probability values, the multiple sub-topics are sorted and recommended to the target object.

Images