Group polarization problem detection method and device for comment area of recommendation system
Through natural language processing and PT-HCL zero-sample position detection model, combining knowledge graph and context information, the shortcomings of group polarization detection in the recommendation system comment area are solved, accurate detection and diversity improvement are achieved, and information cocoons are alleviated and Public opinion fermentation issue.
Patent Information
- Application Number
- CN202410585948.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-13
- Publication Date
- 2024-12-06
AI Technical Summary
The existing technology lacks effective group polarization detection methods and tools in the comment area of the recommendation system, resulting in the fermentation of information cocoons and public opinion. The zero-sample position detection algorithm has a low accuracy rate for neutral position, making it difficult to distinguish between boring and irritating comments, and is easy to form. Misjudgment.
Using natural language processing-based detection tools, through the combination of knowledge graphs, contextual relationship graphs and historical vectors, we can realize the quantification of position detection and polarization degree, establish a PT-HCL zero-sample position detection model, optimize the position detection algorithm, and form a group pole To adjust the early warning and control mechanism, adjust the recommended sorting of the comment area to alleviate polarization.
Accurate detection and quantification of group polarization in the comment area is achieved, the accuracy of position detection is improved, misjudgment is reduced, and the diversity of the comment area is improved through automated early warning and feedback adjustment, and the phenomenon of group polarization is alleviated.
Smart Images

Figure CN119089890A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of recommendation systems, and in particular to a method and device for detecting group polarization problems in a comment area of a recommendation system, and to a method and device for eliminating group polarization based on the group polarization problem detection. Background Art
[0002] Personalized recommendation technology is a tool that helps people obtain interesting content from massive amounts of information, and is also the main technology for personalized recommendation products in information flows. Personalized recommendation technology mainly includes two types: one is based on user history, that is, recommendations are made based on the content that the user has been interested in, and the other is collaborative filtering recommendation, that is, recommendations are made to the current user using other users who are similar to the user. Therefore, if personalized recommendation technology is used improperly, it will inevitably form an information cocoon, especially collaborative filtering recommendation, which is more likely to lead to group polarization.
[0003] Group polarization means that group decisions are more likely to go to extremes than individual decisions, that is, group polarization. Group discussion makes the views of group members more extreme, that is, the originally conservative tend to be more conservative, and the originally adventurous tend to be more adventurous. Group polarization, as an algorithmic chaos, has been studied a lot in the field of public opinion. The concept of group polarization refers to the tendency to extremes in group decision-making, that is, either turning to the adventurous extreme or turning to the conservative extreme. This phenomenon is manifested in specific situations such as "male-female opposition" and "mother-in-law and daughter-in-law opposition" in online public opinion events. If "group polarization" is not controlled, it is easy to cause public opinion to ferment and form a greater and worse impact.
[0004] At present, there are relatively many studies on "group polarization", such as studies from the perspective of social psychology, which are mainly qualitative, or studies from the perspective of social communication, which are mostly quantitative. Research from the perspective of social psychology mainly includes the following aspects: first, qualitative research based on grounded theory, second, empirical research based on user behavior data, and third, network governance based on the emotional dimension. However, research from the perspective of social psychology basically does not consider the role of emotional factors in "group polarization". Therefore, how to accurately identify emotional factors in online public opinion, how to quantify the emotional expression mechanism in the network, and how to construct a general emotional evolution model of network users are the key directions of future research. The conceptual boundaries of research from the perspective of social communication are relatively clear, and research is mostly empirical, including the following aspects: first, personalized recommendation algorithms based on AI technology have become the core content of research; second, the polarization of the population is studied under topics such as politics, social life, and new technologies; third, the information interaction behavior of users in social media, including differences of opinion, likes, comments, and forwarding. Like the research trend from the perspective of social psychology, research from the perspective of social communication does not study the emotional dimension as an important factor. In addition, there are few technical studies on detecting group polarization in the comment section of recommendation systems, and a small amount of detection research is based on graph methods.
[0005] Most of the existing research on the chaos of "group polarization" focuses on the study of mechanisms and theories. The small amount of technical research is also based on graph methods or vector-based quantitative simulation. In real scenarios, especially in the comment area of Internet information flow product recommendation systems, there is a lack of targeted methods to detect the chaos of "group polarization" and tools to measure the degree of "group polarization".
[0006] Existing stance detection methods for comment areas are overly dependent on labeled samples, and some zero-shot stance detection methods lack adaptation to scenarios with high context dependence such as comment areas, and also lack adaptation to the "group polarization" problem. Some zero-shot stance detection algorithms have high accuracy for support and opposition stances, but low accuracy for neutral stances. They also need to be optimized for neutral stances, otherwise it will be difficult to distinguish a large number of boring and spammy comments, which can easily lead to misjudgment.
[0007] In addition, as an algorithmic chaos, the problem of "group polarization" within the personalized recommendation system should establish a reasonable governance mechanism, and relevant algorithmic tools need to be formed to control the content of "group polarization" and deal with related chaos. Summary of the invention
[0008] In view of the deficiencies in the existing technology in studying the "group polarization" problem in the comment section of a personalized recommendation system, and the lack of a technical solution that can provide feedback and adjustment for the "group polarization" problem in the comment section, the present invention proposes a method and device for detecting the group polarization problem in the comment section of a recommendation system. The method establishes a detection tool based on a natural language processing method, realizes the quantification of the degree of "group polarization", forms a feedback and adjustment mechanism for the "group polarization" content, and realizes the technical implementation of the governance of the "group polarization" chaos.
[0009] like Figure 4 As shown, the present invention proposes a method for detecting group polarization problems in a comment area of a recommendation system, comprising:
[0010] The stance detection step S1 extracts the topic viewpoints of the main comment post specified by the recommendation system to obtain the viewpoints of the main post; performs stance detection on the comment area of the main comment post to obtain the stance categories of each comment in the comment area relative to the viewpoint of the main post;
[0011] Part of speech detection step S2, by extracting keywords from each comment, obtain the intensity of the stance category of each comment; sort the comments in the comment area according to the comment time, count the number of comments in each stance category in the initial time period of the comments, and obtain the average polarization degree of each stance category according to the intensity of each comment in the initial time period as a benchmark;
[0012] Polarization alarm step S3, continuously extracting subsequent time periods of the same length as the initial time period but later in time, calculating the average polarization degree of each position category in the subsequent time period, and comparing it with the benchmark to obtain the polarization deviation degree of the position, and alarming when the polarization deviation degree exceeds the threshold.
[0013] In the method for detecting group polarization problems in the comment area of a recommendation system, the stance detection step includes:
[0014] The comment text in the comment area is expanded through the knowledge graph to obtain the extended enhanced text, and the extended enhanced text is used to perform stance detection to form a knowledge graph enhanced vector;
[0015] Convert the account history text of the comment text into a history vector;
[0016] Obtaining a contextual comment reply of the comment text, establishing a contextual relationship map between the comment text and the contextual comment reply, and enhancing the comment text through the contextual relationship map to obtain a contextual enhancement vector;
[0017] The context enhancement vector, the knowledge graph enhancement vector and the history vector are combined to obtain a stance discriminant vector, and stance detection is performed on the stance discriminant vector to generate a stance category of the comment text.
[0018] The method for detecting group polarization problems in the comment area of the recommendation system, wherein the polarization alarm step comprises:
[0019] When the polarization deviation exceeds the threshold, the polarization deviation is subtracted from 1 to obtain a polarization adjustment parameter, and the polarization adjustment parameter is used to recommend comments that are opposite to the position of the users who have made comments in the comment area.
[0020] The method for detecting group polarization problems in the comment area of the recommendation system, wherein a PT-HCL zero-sample stance detection model based on hierarchical contrastive learning is trained using samples that conform to the stance distribution of the comment area to obtain a stance detection model, stance detection is performed on the extended enhanced text through the stance detection model, and the account history text of the comment text is converted into a history vector through the stance detection model.
[0021] like Figure 5 As shown, the present invention proposes a group polarization problem detection device B in the comment area of a recommendation system, which includes:
[0022] The stance detection module M1 extracts the topic viewpoints of the main comment post specified by the recommendation system to obtain the main post viewpoints; performs stance detection on the comment area of the main comment post to obtain the stance categories of each comment in the comment area relative to the main post viewpoints;
[0023] The part-of-speech detection module M2 extracts the keywords of each comment to obtain the intensity of the position category of each comment; sorts the comments in the comment area according to the comment time, counts the number of comments of each position category in the initial time period of the comments, and obtains the average polarization degree of each position category according to the intensity of each comment in the initial time period as a benchmark;
[0024] The polarization alarm module M3 continuously extracts subsequent time periods of the same length as the initial time period but later in time, calculates the average polarization degree of each position category in the subsequent time period, and compares it with the benchmark to obtain the polarization deviation degree of the position, and issues an alarm when the polarization deviation degree exceeds a threshold.
[0025] The device for detecting group polarization problems in the comment area of the recommendation system, wherein the stance detection in the stance detection module includes:
[0026] The comment text in the comment area is expanded through the knowledge graph to obtain the extended enhanced text, and the extended enhanced text is used to perform stance detection to form a knowledge graph enhanced vector;
[0027] Convert the account history text of the comment text into a history vector;
[0028] Obtaining a contextual comment reply of the comment text, establishing a contextual relationship map between the comment text and the contextual comment reply, and enhancing the comment text through the contextual relationship map to obtain a contextual enhancement vector;
[0029] The context enhancement vector, the knowledge graph enhancement vector and the history vector are combined to obtain a stance discriminant vector, and stance detection is performed on the stance discriminant vector to generate a stance category of the comment text.
[0030] The device for detecting group polarization problems in the comment area of the recommendation system, wherein the polarization alarm module comprises:
[0031] When the polarization deviation exceeds the threshold, the polarization deviation is subtracted from 1 to obtain a polarization adjustment parameter, and the polarization adjustment parameter is used to recommend comments that are opposite to the position of the users who have made comments in the comment area.
[0032] The device for detecting group polarization problems in the comment area of the recommendation system, wherein a PT-HCL zero-sample stance detection model based on hierarchical contrastive learning is trained using samples that conform to the stance distribution of the comment area to obtain a stance detection model, and stance detection is performed on the extended enhanced text through the stance detection model, and the account history text of the comment text is converted into a history vector through the stance detection model.
[0033] The present invention provides an electronic device, which includes the device for detecting group polarization problems in a comment area of a recommendation system.
[0034] The electronic device is connected to an information display device, and the information display device is used to display the degree of polarization deviation using display parameters and attributes set by a user or through an artificial intelligence model.
[0035] The present invention proposes a storage medium for storing a computer program for executing the method for detecting group polarization problems in a comment area of a recommendation system.
[0036] It can be seen from the above scheme that the advantages of the present invention are:
[0037] Compared with existing methods, this method can realize a set of technical detection tools based on natural language processing and more targeted for the problem of "group polarization" in the comment area. It can not only instantly detect whether the chaos of "group polarization" occurs, but also characterize the degree of polarization, providing a basis for subsequent early warning and control. The present invention also implements a set of stance detection algorithms with better detection effects for the problem of "group polarization" in the comment area. The algorithm can more accurately identify the support, opposition, and neutral stances in the comment area, and further improve the recognition accuracy of the chaos of "group polarization". The present invention also proposes a "recommendation system group polarization early warning and control mechanism", which can automatically and in real time warn and control the chaos through algorithms, feedback and adjust the recommendation system, improve the diversity of the comment area, and alleviate "group polarization". BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 A flowchart for performing polarization detection on the comment area;
[0039] Figure 2 Diagram of the complete stance detection process for the review text;
[0040] Figure 3 A diagram of the recommended sorting strategy for adjusting the comment area according to the degree of polarization deviation;
[0041] Figure 4 is a flow chart of the method of the present invention;
[0042] Figure 5 It is a module diagram of the device of the present invention;
[0043] Figure 6 This is a schematic diagram of the structure of a first electronic device of the present invention;
[0044] Figure 7 This is a schematic diagram of the application environment structure of the first electronic device of the present invention;
[0045] Figure 8 It is a schematic diagram of the structure of a second electronic device of the present invention.
[0046] Reference numerals:
[0047] 1- Extract the main post topic viewpoints;
[0048] 2-Optimized stance detection in the comment area scenario;
[0049] 3-Stance strength detection combining word stance and part of speech;
[0050] 4-Measurement of average group polarization;
[0051] 5-Comment text;
[0052] 6-Account history text;
[0053] 7- Knowledge graph expansion;
[0054] 8-Zero-sample stance detection algorithm model;
[0055] 9-Standpoint sample distribution optimization;
[0056] 10-Preliminary position results;
[0057] 11-Comment position transfer relationship diagram;
[0058] 12-Knowledge graph enhancement vector;
[0059] 13-GCN context enhancement vector;
[0060] 14-historical information enhancement vector;
[0061] 15-Multiple enhanced vector fusion;
[0062] 16-Fully connected correction outputs the final position;
[0063] 17-Strategy for breaking up comment ranking in personalized recommendation system;
[0064] 18-Comment area group polarization detection warning;
[0065] 19-Generation of polarization degree adjustment parameters;
[0066] S1-Stance detection step;
[0067] S2-part-of-speech detection step;
[0068] S3-polarization alarm step;
[0069] M1-Stance detection module;
[0070] M2-part-of-speech detection module;
[0071] M3-Polarization alarm module;
[0072] A-First electronic device;
[0073] B- Group polarization problem detection device in the comment area of the recommendation system;
[0074] C-data acquisition equipment;
[0075] D-information display device;
[0076] 1000 - second electronic device;
[0077] Ⅰ-computational unit;
[0078] II-ROM;
[0079] III-RAM;
[0080] IV-bus;
[0081] V-interface;
[0082] VI - input unit;
[0083] VII-output unit;
[0084] VIII- Storage medium;
[0085] Ⅸ-Communication unit. DETAILED DESCRIPTION
[0086] It should be noted that the processor described in the present invention is the control center of the electronic device, which can be a processor or a general term for multiple processing elements. For example, it can be one or more central processing units (CPUs), or application specific integrated circuits (ASICs), or one or more integrated circuits configured to implement the embodiments of the present invention, such as one or more microprocessors (digital signal processors, DSPs), or one or more field programmable gate arrays (FPGAs).
[0087] Optionally, the processor can perform various functions of the electronic device by running or executing a software program stored in the memory, and calling data stored in the memory.
[0088] In a specific implementation, as an embodiment, the processor may include one or more CPUs. Each of these processors may be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). The processor here may refer to one or more devices, circuits, and / or processing cores for processing data (such as computer program instructions). Electronic devices may include: servers, desktop computers, laptops, smart phones, tablet computers, embedded computers, etc., wherein the embedded computers include vehicles and robots, etc.
[0089] The memory is used to store the software program for executing the solution of the present invention, and the execution is controlled by the processor. The specific implementation method can refer to the above method embodiment, which will not be repeated here.
[0090] It should be noted that the structure of the electronic device shown in the drawings of the present invention does not constitute a limitation thereto, and the actual knowledge structure recognition device may include more or fewer components than shown in the drawings, or a combination of certain components, or a different arrangement of components.
[0091] The above embodiments can be implemented in whole or in part by software, hardware (such as circuits), firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.
[0092] It should also be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.
[0093] In the present invention, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.
[0094] It should also be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.
[0095] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, 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 devices or units, which can be electrical, mechanical or other forms.
[0096] 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.
[0097] 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.
[0098] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.
[0099] When the inventor was conducting research on the chaos of "group polarization", he found that most of the research on the chaos of "group polarization" was an analysis of the mechanism of chaos generation and the impact of public opinion, and there was less research on detection technology for discovering group polarization chaos. Research on "group polarization" under recommendation systems mostly used simulation technology and graph-based methods. In the comment area scenario, there is a lack of targeted natural language processing methods to detect the chaos of "group polarization"; the inventor also found that although zero-sample stance detection is more suitable for detecting opinions in the comment area, its detection effect on different stances varies greatly, and it lacks correlation adaptation to the context of the comment area and usage adaptation under the problem of "group polarization"; in addition, there is a lack of specific regulatory mechanisms for this chaos in the comment area, and a lack of reasonable control of the "group polarization" content through technical means.
[0100] After an in-depth study of the chaos of "group polarization", the inventors found that to solve the above problems, we must first combine natural language processing methods to propose a "group polarization" detection tool for the comment area of the recommendation system, which can not only detect whether "group polarization" occurs, but also characterize the severity of "group polarization" to facilitate monitoring and early warning; secondly, we must optimize the stance detection method, improve and adapt the stance detection algorithm according to the comment area scenario and the "group polarization" problem to achieve better detection and identification effects; finally, we must combine the mechanism of the recommendation system, and make reasonable feedback adjustments through the detected "group polarization" phenomenon and the degree of polarization, so as to realize the automatic elimination of this chaos in the recommendation system, rather than aggravating this phenomenon, so as to protect the network environment and enhance the user experience.
[0101] In order to make the above features and effects of the present invention more clearly and understandably described, embodiments are given below and described in detail with reference to the accompanying drawings. This specification discloses one or more embodiments that include the features of the present invention. The disclosed embodiments are only for illustration. The scope of protection of the present invention is not limited to the disclosed embodiments, and the present invention is defined by the attached claims.
[0102] The present invention proposes a method and device for detecting group polarization problems in a recommendation system comment area, comprising the following steps:
[0103] Step 1: Detect the “group polarization” phenomenon in the comment area. Figure 1As shown in the figure, first, it is necessary to extract topic viewpoints for the main post of the comment. For example, for topics related to "gender confrontation", if the main post is "Men are all ordinary and confident", it can be clearly seen that the main post holds the view of "supporting women and opposing men". Then, for the replies to the main post in the subsequent comments, they will have opposite views. Secondly, use the optimized stance detection algorithm for the content in the comment area to detect the stance of comment sentences, forming a stance classification of supporting, neutral, and opposing the main post. Thirdly, introduce word stance and part-of-speech detection to measure the degree of the expressed stance. For example, derogatory words such as "scumbag" itself contain content with a strong opposing stance towards men, as well as the use of degree adverbs such as "very" and "extremely". The specific intensity can be obtained by looking up a table. The table has the intensity weights corresponding to the keywords. By classifying the intensity of common words (keywords), for example, positive words, negative words, neutral words, strong positive words, and strong negative words, to form this table. Finally, calculate the average polarization degree index, that is, sort the comment area according to time, detect and count the two polarization stances of support and opposition for an initial time period in the comments, and use the polarization degree to calculate the average polarization degree of support and opposition weighted, and use this as a control benchmark. Then continuously extract different time periods with the same length but later in time for the detection and statistics of the two polarization stances of support and opposition, calculate the average polarization degree of support and opposition, compare it with the benchmark, calculate the polarization deviation degree of the support or opposition stance, and define that the polarization degree of the support or opposition stance deviates from the benchmark by 30% as the existence of group polarization chaos and include it in the monitoring and early warning.
[0104] It should be noted that polarization may not necessarily not occur in the comments during the initial time period. Counting support and opposition can measure the initial polarization degree. The control benchmark is because after considering the influence of time fermentation, whether the polarization degree further increases. At this time, comparing with the initial state, the initial state is the benchmark.
[0105] Step 2: Adapt and optimize the stance detection algorithm for the comment area scenario. Such as Figure 2 shown Figure 2 This is an example of the process of applying this stance detection algorithm in the comment area. Figure 2 The process should correspond to the optimized stance detection algorithm.
[0106] Step 2 is actually a process of judging the stance of an account's comment. General stance detection is pure text detection, that is, inputting the text of the comment area, and the stance detection model gives the stances of support, opposition, and neutral. Since the comment text is short, relying solely on the text will lack information, such as missing context information, etc. Therefore, the process of Step 2 is a process of supplementing relevant information.
[0107] The complete stance detection process of a comment text is as follows: Taking the topic of male-female opposition as an example, first of all, the text itself is used to determine whether there are some aliases for nouns, such as "double doors", which generally refers to men but is not directly involved in the text. Graph enhancement is required. For example, the sentence "If it is a double door, I support it". After knowledge graph enhancement, it becomes "If it is a double door (male) I support it". The enhanced content is used for stance detection to form a knowledge graph enhancement vector. At the same time, the historical texts of the account that publishes comments, such as "favorite husband" and "support gender equality", can be converted into historical vectors through the algorithm model. Another point is unique to the comment area, that is, the context of the comment area. The comment reply has a premise. The text indicates "I oppose", and the reply itself may be "support men", so it becomes "oppose men" after the context is combined. Therefore, through the relationship graph of the context comment reply, the current text is enhanced through the context, such as "I oppose 'support men'", and the context enhancement vector is formed at this time. The context enhancement vector, the knowledge graph enhancement vector, and the historical vector are combined to form the final stance discrimination vector, and then the model is discriminated to generate the accurate stance of the text itself.
[0108] First, we select the PT-HCL zero-sample stance detection algorithm based on hierarchical contrastive learning to enhance the algorithm’s adaptability to different detection topics. At the same time, the biggest problem with this algorithm is that it has a low accuracy rate for neutral stance detection, which does not match the scenario where there is a large amount of neutral spam in the comment area. Therefore, we need to retrain the algorithm model with samples that match the stance distribution of the comment area to adapt to the comment area scenario. Secondly, the content of the comment area is short and lacks background knowledge. Introducing knowledge graph information can significantly enhance the effect of stance detection. Therefore, we extract entity words and expand related vocabulary through the knowledge graph to enhance word vectors. ; Secondly, the comment scenario is a scenario with strong contextual association and the comment stance is transitive, so the conversation thread in the comment area is converted into a graph structure and the GCN method is used for content enhancement. That is, A supports B, B supports C, then A and C also establish an edge to form a graph structure. Therefore, after the zero-sample detection algorithm forms a vector expression, the vector can be corrected according to the graph relationship in the comment area; in addition, the content containing the account's historical posts can represent the account's consistent stance, and information enhancement based on the content of historical posts can improve the accuracy of stance detection; finally, the vectors of multiple rounds of enhancement are fused to determine the final stance.
[0109] Step 3: Establish a feedback adjustment mechanism based on the degree of polarization deviation of the support or opposition positions. Figure 3As shown, the recommendation sorting strategy of the comment area is adjusted according to the degree of polarization deviation. For male users, more recommendations are given to support female positions, and for female users, more recommendations are given to support male positions. The head content of the recommendation list is broken up according to the degree of polarization. The higher the degree of polarization, the greater the degree of dispersion, so that users are exposed to different voices instead of repeatedly reinforcing the same point of view, which leads to further polarization.
[0110] At the end of step 1, the polarization deviation value is obtained by comparing with the benchmark. For example, if the polarization deviation is 40%, which exceeds the defined threshold of 30%, the recommendation sorting logic needs to be adjusted to downgrade certain polarized viewpoints. For example, if men tend to support men and the polarization deviation is 40%, then (1-40%), or 0.6 times, is used to downgrade the content that supports men's views. After the downgrade, the views that support women will be recommended to men first. The higher the deviation, the greater the downgrade coefficient, which is used to feedback and adjust the diversity sorting and fragmentation strategy in the recommendation distribution.
[0111] The specific process of diversity disruption is as follows: polarization means that men see more content that supports men, for example, 90% support men, while women see more content that supports women, for example, 90% support women; diversity disruption means adjusting the ratio of content supporting men and women. The ideal state is that men see 50% support men and 50% support women. Disruption means balancing the recommendation ratios of different viewpoints.
[0112] The following is a system embodiment corresponding to the above method embodiment. This embodiment can be implemented in conjunction with the above embodiment. The relevant technical details mentioned in the above embodiment are still valid in this embodiment. In order to reduce repetition, they are not repeated here. Accordingly, the relevant technical details mentioned in this embodiment can also be applied in the above embodiment.
[0113] like Figure 5 As shown, the present invention proposes a group polarization problem detection device B in the comment area of a recommendation system, which includes:
[0114] The stance detection module M1 extracts the topic viewpoints of the main comment post specified by the recommendation system to obtain the main post viewpoints; performs stance detection on the comment area of the main comment post to obtain the stance categories of each comment in the comment area relative to the main post viewpoints;
[0115] The part-of-speech detection module M2 extracts the keywords of each comment to obtain the intensity of the position category of each comment; sorts the comments in the comment area according to the comment time, counts the number of comments of each position category in the initial time period of the comments, and obtains the average polarization degree of each position category according to the intensity of each comment in the initial time period as a benchmark;
[0116] The polarization alarm module M3 continuously extracts subsequent time periods of the same length as the initial time period but later in time, calculates the average polarization degree of each position category in the subsequent time period, and compares it with the benchmark to obtain the polarization deviation degree of the position, and issues an alarm when the polarization deviation degree exceeds a threshold.
[0117] The device for detecting group polarization problems in the comment area of the recommendation system, wherein the stance detection in the stance detection module includes:
[0118] The comment text in the comment area is expanded through the knowledge graph to obtain the extended enhanced text, and the extended enhanced text is used to perform stance detection to form a knowledge graph enhanced vector;
[0119] Convert the account history text of the comment text into a history vector;
[0120] Obtaining a contextual comment reply of the comment text, establishing a contextual relationship map between the comment text and the contextual comment reply, and enhancing the comment text through the contextual relationship map to obtain a contextual enhancement vector;
[0121] The context enhancement vector, the knowledge graph enhancement vector and the history vector are combined to obtain a stance discriminant vector, and stance detection is performed on the stance discriminant vector to generate a stance category of the comment text.
[0122] The device for detecting group polarization problems in the comment area of the recommendation system, wherein the polarization alarm module comprises:
[0123] When the polarization deviation exceeds the threshold, the polarization deviation is subtracted from 1 to obtain a polarization adjustment parameter, and the polarization adjustment parameter is used to recommend comments that are opposite to the position of the users who have made comments in the comment area.
[0124] The device for detecting group polarization problems in the comment area of the recommendation system, wherein a PT-HCL zero-sample stance detection model based on hierarchical contrastive learning is trained using samples that conform to the stance distribution of the comment area to obtain a stance detection model, and stance detection is performed on the extended enhanced text through the stance detection model, and the account history text of the comment text is converted into a history vector through the stance detection model.
[0125] like Figure 6 As shown, the present invention further proposes a first electronic device A in another embodiment, which includes the group polarization problem detection device in the comment area of the recommendation system.
[0126] like Figure 7As shown, the first electronic device A can also be connected to the data acquisition device C and the information display device D through a wired or wireless information transmission scheme. The data acquisition device C is used to collect and obtain the text to be identified and classified, such as the comment main post text and the comment area text described in the embodiment of the present invention. The information display device D is used to display the degree of polarization deviation obtained by the analysis of the present invention.
[0127] The information display device D can process the data output by the first electronic device A based on the information display mechanism to improve the readability of the data output by the first electronic device A. The information display mechanism can be manually preset, for example, the data output by the first electronic device A is visually displayed, which can be based on the display parameters and / or attributes set by the user. The display parameters can be, for example, the display data range, and the display attributes can be, for example, the display font, color, whether to scroll and play, etc. The user is presented with the key information specified by the user, such as the degree of polarization deviation, which comments are the most extreme, etc. The user can understand this information more timely without having to access the secondary page or scroll the page, saving the user's operation. Or the information display mechanism can be an artificial intelligence AI display model, which can learn the user's key information based on the user's previous usage habits, such as viewing time, number of clicks, number of edits, etc., and then automatically present rich and necessary key information to the user.
[0128] The present invention also proposes a storage medium VIII in another embodiment for storing a computer program for executing the method for detecting group polarization problems in the comment area of the recommendation system. It should be understood that the storage medium in the embodiment of the present invention may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct RAM bus RAM (DR RAM).
[0129] Figure 8 A schematic block diagram of a second electronic device 1000 that can be used to implement an embodiment of the present invention is shown. The second electronic device 1000 electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The second electronic device 1000 can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples, and are not intended to limit the implementation of the present invention described and / or required herein. The second electronic device 1000 may be the same or different from the first electronic device A.
[0130] The second electronic device 1000 includes a computing unit I, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory II (ROM) or a computer program loaded from a storage medium VIII into a random access memory (RAM) III. In RAM III, various programs and data required for the operation of the device 1000 can also be stored. The computing unit I, ROM II, and RAM III are connected to each other via a bus IV. An input / output (I / O) interface V is also connected to the bus IV.
[0131] A plurality of components in the second electronic device 1000 are connected to the I / O interface V, including: an input unit VI, such as a keyboard, a mouse, etc.; an output unit VII, such as various types of displays, speakers, etc.; a storage medium VIII, such as a disk, an optical disk, etc.; and a communication unit IX, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit IX allows the second electronic device 1000 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.
[0132] The computing unit I may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of computing unit I include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit I performs the various methods and processes described above, such as method steps S1-S3. For example, in some embodiments, the method may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage medium VIII. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 1000 via ROM II and / or communication unit IX. When the computer program is loaded into RAM III and executed by the computing unit I, one or more steps of the method described above may be performed. Alternatively, in other embodiments, the computing unit I may be configured to execute the method in any other appropriate manner (e.g., by means of firmware).
[0133] In summary, the present invention proposes a real-time detection technology for the "group polarization" chaos in the comment area of the recommendation system, proposes an evaluation index for describing "group polarization", and realizes the quantification of the degree of polarization; forms an algorithm detection tool for the "group polarization" chaos to realize automatic judgment. The present invention proposes a scene-adapted stance detection technology, which uses knowledge graph, comment area context, and user historical information context to enhance the zero-sample stance detection algorithm, and optimizes the detection effect of the algorithm on the neutral stance to achieve better stance detection effect for the "group polarization" problem in the comment area; forms a stance detection tool for the comment area scene, and is more suitable for detecting the "group polarization" problem, and the detection accuracy is significantly improved. Based on the detection of "group polarization", the present invention proposes a "recommendation system group polarization early warning and control mechanism", which improves the diversity of the comment area and alleviates "group polarization" by describing the degree of polarization in the comment area, giving immediate early warnings, and feedback-adjusting the sorting strategy of the comment area of the recommendation system; the mechanism is very compatible with the real-time "group polarization" detection technology, and can feedback-adjust the recommendation system according to the detected results to achieve chaos management.
[0134] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and the implementation modes, and they can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and the illustrations shown and described herein.
Claims
1. A method for detecting group polarization problems in the comment area of a recommendation system, characterized in that: include: The stance detection step extracts the topic viewpoints of the main comment posts specified by the recommendation system to obtain the main post viewpoints; Perform stance detection on the comment area of the main post to obtain the stance category of each comment in the comment area relative to the main post; In the part-of-speech detection step, the keywords of each comment are extracted to obtain the intensity of the position category of each comment; the comments in the comment area are sorted according to the comment time, and the number of comments in each position category in the initial time period is counted. Based on the intensity of each comment in the initial time period, the average polarization degree of each position category is obtained as a benchmark; The polarization alarm step continuously extracts subsequent time periods of the same length as the initial time period but later in time, calculates the average polarization degree of each position category in the subsequent time period, and compares it with the benchmark to obtain the polarization deviation degree of the position. When the polarization deviation degree exceeds the threshold, an alarm is issued.
2. The method for detecting group polarization problems in a recommendation system comment area according to claim 1, characterized in that: The stance detection step includes: The comment text in the comment area is expanded through the knowledge graph to obtain the extended enhanced text, and the extended enhanced text is used to perform stance detection to form a knowledge graph enhanced vector; Convert the account history text of the comment text into a history vector; Obtaining a contextual comment reply of the comment text, establishing a contextual relationship map between the comment text and the contextual comment reply, and enhancing the comment text through the contextual relationship map to obtain a contextual enhancement vector; The context enhancement vector, the knowledge graph enhancement vector and the history vector are combined to obtain a stance discriminant vector, and stance detection is performed on the stance discriminant vector to generate a stance category of the comment text.
3. The method for detecting group polarization problems in a recommendation system comment area according to claim 1 or 2, characterized in that: The polarization alarm steps include: When the polarization deviation exceeds the threshold, the polarization deviation is subtracted from 1 to obtain a polarization adjustment parameter, and the polarization adjustment parameter is used to recommend comments that are opposite to the position of the users who have made comments in the comment area.
4. The method for detecting group polarization problems in a recommendation system comment area according to claim 2, characterized in that: The samples that match the stance distribution of the comment area are used to train the PT-HCL zero-sample stance detection model based on hierarchical contrastive learning to obtain a stance detection model. The stance detection model is used to perform stance detection on the extended enhanced text, and the account history text of the comment text is converted into a history vector through the stance detection model.
5. A device for detecting group polarization problems in a recommendation system comment area, characterized in that: include: The stance detection module extracts the topic opinions of the main comment posts specified by the recommendation system and obtains the opinions of the main posts; Perform stance detection on the comment area of the main post to obtain the stance category of each comment in the comment area relative to the main post; The part-of-speech detection module extracts the keywords of each comment and obtains the intensity of the stance category of each comment; Sort the comments in the comment area by comment time, count the number of comments in each position category in the initial comment period, and obtain the average polarization degree of each position category based on the intensity of each comment in the initial period as a benchmark; The polarization alarm module continuously extracts subsequent time periods of the same length as the initial time period but later in time, calculates the average polarization degree of each position category in the subsequent time period, and compares it with the benchmark to obtain the polarization deviation degree of the position. When the polarization deviation degree exceeds the threshold, an alarm is issued.
6. The device for detecting group polarization problems in a recommendation system comment area according to claim 5, characterized in that: The stance detection in the stance detection module includes: The comment text in the comment area is expanded through the knowledge graph to obtain the extended enhanced text, and the extended enhanced text is used to perform stance detection to form a knowledge graph enhanced vector; Convert the account history text of the comment text into a history vector; Obtaining a contextual comment reply of the comment text, establishing a contextual relationship map between the comment text and the contextual comment reply, and enhancing the comment text through the contextual relationship map to obtain a contextual enhancement vector; The context enhancement vector, the knowledge graph enhancement vector and the history vector are combined to obtain a stance discriminant vector, and stance detection is performed on the stance discriminant vector to generate a stance category of the comment text.
7. The device for detecting group polarization problems in a comment area of a recommendation system according to claim 5 or 6, characterized in that: The polarization alarm module includes: When the polarization deviation exceeds the threshold, the polarization deviation is subtracted from 1 to obtain a polarization adjustment parameter, and the polarization adjustment parameter is used to recommend comments that are opposite to the position of the users who have made comments in the comment area.
8. The device for detecting group polarization problems in a recommendation system comment area according to claim 6, characterized in that: The samples that match the stance distribution of the comment area are used to train the PT-HCL zero-sample stance detection model based on hierarchical contrastive learning to obtain a stance detection model. The stance detection model is used to perform stance detection on the extended enhanced text, and the account history text of the comment text is converted into a history vector through the stance detection model.
9. An electronic device, characterized in that: The invention comprises a group polarization problem detection device in a comment area of a recommendation system as described in claims 5-8.
10. The electronic device according to claim 9, characterized in that: The electronic device is connected to an information display device, which is used to display the degree of polarization deviation using display parameters and attributes set by a user or through an artificial intelligence model.
11. A storage medium for storing a computer program for executing the method for detecting group polarization problems in a comment area of a recommendation system according to claims 1-4.
