A method and device for detecting diversity of recommendation results of a recommendation system

Abstract

This application discloses a method and device for detecting the diversity of recommendation results of a recommendation system, comprising: obtaining the recommendation results of the recommendation system and classifying the recommendation results based on a preset topic classification model to obtain the topic categories of the recommended content; determining similar topic content within the topic categories of the recommended content, and performing emotional stance detection using a preset emotional stance detection model; and calculating a diversity evaluation index that considers emotional stance diversity based on the emotional stance detection results. The method of this application incorporates the diversity of the emotional stance dimension into the recommendation system diversity index, thereby proposing a recommendation system diversity detection method that considers emotional stance diversity.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method and device for detecting the diversity of recommendation results of a recommendation system. Background Art

[0002] The widespread application of recommendation algorithms based on artificial intelligence technology on Internet platforms has greatly improved the efficiency of information production and dissemination, and brought about profound changes in the way and activities of dissemination. Common recommendation systems in our lives are mainly used in search engines (such as Baidu, Google) and platform recommendation mechanisms (such as Toutiao, Douyin). The recommendation algorithms currently used on platforms are more likely to be based on the content and tags visited by users and their corresponding clicks, favorites, likes, comments, etc.

[0003] Build models to recommend personalized content to different users. While this approach makes information access more convenient, it also presents some challenges. Traditionally, evaluating recommendation system performance based on accuracy metrics can lead to overfitting, where, for example, the model is overfitted to the point where it cannot detect any other content of interest. Over time, focusing solely on accuracy without considering diversity can lead to the phenomenon of "information cocoons." Therefore, measuring the diversity of recommendation systems has become an urgent issue, and much research has been conducted based on this.

[0004] Existing methods generally distinguish between two dimensions of diversity in recommender systems: individual-level diversity, which focuses on avoiding recommending redundant (but still relevant) items to users given a previously recommended list. System-level diversity, on the other hand, focuses on the system's ability to recommend less popular or hard-to-find items. Following these two dimensions, specific metrics for measuring recommender system diversity are further categorized into two types: distance-based metrics assess diversity by calculating pairwise distances between all items in a recommendation list, with smaller distance values ​​indicating poorer diversity. Coverage-based metrics, on the other hand, aim to quantify the breadth of topics covered in a given list of recommendations, primarily examining topic diversity. However, neither diversity metric emphasizes diversity in the sentiment dimension, which can still lead to "emotionally monotonous" or "emotionally polarized" recommendations.

[0005] All existing metrics focus primarily on the diversity of the recommended content itself, such as category and topic diversity and distance measurement, but none consider diversity in sentiment. For example, a system may recommend many content categories, but for content in the same category, it only recommends positive and proactive evaluations, and not neutral or negative evaluations. While such a recommendation list has high topic diversity, it still creates a "polarization phenomenon" and lacks good diversity. This is especially true on news recommendation platforms. If users can only browse information in a single field for a long time, they will easily lose their comprehensive understanding of the facts and their objective evaluation, resulting in "truth scarcity" and "emotional polarization." Therefore, it is necessary to factor diversity in sentiment into the metrics. Summary of the Invention

[0006] The present invention provides a method and device for detecting the diversity of recommendation results in a recommendation system, which is used to incorporate the diversity of the emotional and stance dimensions into the recommendation system diversity index. A recommendation system diversity detection method that takes into account the diversity of emotional and stance dimensions is proposed.

[0007] This embodiment of the present application provides a method for detecting the diversity of recommendation results of a recommendation system, including:

[0008] Obtaining recommendation results from the recommendation system, and classifying the recommendation results based on a preset topic classification model to obtain topic categories of recommended content;

[0009] Determine similar topics in the recommended content category and perform sentiment detection using a preset sentiment detection model.

[0010] According to the results of emotional stance detection, the diversity evaluation index considering the diversity of emotional stance is calculated.

[0011] Optionally, obtaining the recommendation result of the recommendation system includes:

[0012] For a single person, obtain personal user information and personal recommendation results;

[0013] For multiple people, multiple user information and recommendation results of the recommendation system for multiple people are obtained, and the multiple recommendation results are aggregated.

[0014] Optionally, performing emotional stance detection using a preset emotional stance detection model includes:

[0015] Emotional stance detection is performed using a preset emotional stance detection model to classify emotional stances into positive, neutral, and negative.

[0016] Optionally, based on the emotion stance detection results, the diversity evaluation indicators that consider the diversity of emotion stances are calculated, including:

[0017] For a single person, the diversity evaluation index considering the diversity of emotional stances is calculated to meet the following requirements:

[0018]

[0019] Among them, n s It is expressed as the number of different topics, l is the list of items under consideration, i is the i-th topic, j is the j-th sentiment attitude, o is the sentiment attitude, c(l,s i , o j ) indicates that the list l belongs to the subject s i And the attitude is o j The number of items.

[0020] Optionally, based on the emotion stance detection results, the diversity evaluation indicators that consider the diversity of emotion stances are calculated, including:

[0021] For several people, the diversity evaluation index considering the diversity of emotional stances satisfies:

[0022]

[0023] Among them, n s It is expressed as the number of different topics, k represents the kth sentiment attitude, j represents the jth sentiment attitude, e i，j It is expressed as the number of occurrences of items with item topic i and sentiment stance j, e i，k It is expressed as the number of occurrences of items whose topic is i and whose sentiment stance is k.

[0024] An embodiment of the present application further proposes a device for detecting the diversity of recommendation results of a recommendation system, comprising a processor and a memory, wherein a computer program is stored on the memory, and when the computer program is executed by the processor, the steps of the method for detecting the diversity of recommendation results of a recommendation system as described above are implemented.

[0025] An embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the aforementioned method for detecting the diversity of recommendation results of a recommendation system are implemented.

[0026] The embodiment of the present application proposes a recommendation system diversity detection method that takes into account the diversity of emotional stance dimensions by integrating the diversity of the recommendation system diversity index.

[0027] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present application. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0029] Figure 1 This is a basic flow chart of a method for detecting the diversity of recommendation results of a recommendation system according to an embodiment of the present application;

[0030] Figure 2 The following is a schematic diagram of the overall process of the method for detecting the diversity of recommendation results of the recommendation system according to an embodiment of the present application. DETAILED DESCRIPTION

[0031] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0032] Existing distance-based metrics primarily use the Intra-List Average Distance (ILMD) to capture recommendation list diversity. This formula measures the average distance between pairs of items in a list, with smaller distances indicating lower diversity. Distance measurements primarily use cosine similarity, Euclidean distance, and Jaccard distance.

[0033] Another existing coverage-based metric is Catalog Coverage (C-Coverage) and Subtopic Coverage (S-Coverage). C-Coverage primarily measures the ratio of the number of items recommended to a user to the total number of items. S-Coverage primarily measures the ratio of the number of item categories recommended to a user to the total number of item categories.

[0034] Regardless of which measurement indicator is used, they mainly focus on the diversity of the recommended content itself, such as category topic diversity and distance measurement, but none of them takes into account the diversity of emotional stance.

[0035] Based on this, the embodiment of the present application provides a method for detecting the diversity of recommendation results of a recommendation system, such as Figure 1 、 Figure 2 As shown, the following steps are included:

[0036] In step S101, the recommendation results of the recommendation system are obtained, and the recommendation results are classified based on a preset topic classification model to obtain the topic category of the recommended content. In some embodiments, obtaining the recommendation results of the recommendation system includes:

[0037] For a single person, obtain personal user information and personal recommendation results;

[0038] For multiple people, multiple user information and recommendation results of the recommendation system for multiple people are obtained, and the multiple recommendation results are aggregated.

[0039] In some specific examples, the following process is used for individual-level diversity assessment:

[0040] Obtain specific user information and recommendation results. For a given recommendation list (including recommendation results), pass the items through a topic classification model, such as the LDA model, to extract and record the topic types to which the items belong.

[0041] The following process is used for system-level diversity evaluation:

[0042] Obtain multiple specific user information and recommendation results, aggregate multiple recommendation result lists, and for a given recommendation list (aggregated result), pass the items through a topic classification model, such as the LDA model, to extract and record the topic type to which the items belong.

[0043] In step S102, similar subject content in the subject category of the recommended content is determined, and emotional stance detection is performed using a preset emotional stance detection model. In some embodiments, performing emotional stance detection using a preset emotional stance detection model includes:

[0044] Sentiment detection is performed using a preset sentiment detection model to categorize sentiment into positive, neutral, and negative. In some specific applications, after the topic model is applied, similar topic recommendation items are tested using a sentiment classification model, and sentiment is categorized into positive, neutral, and negative, with the data recorded.

[0045] In step S103, based on the emotion stance detection results, a diversity evaluation index that considers the diversity of emotion stances is calculated. Specifically, the recorded data is subjected to SD Index calculation to obtain a diversity evaluation score for the recommendation list.

[0046] In some embodiments, based on the emotion stance detection results, calculating a diversity evaluation index that considers the diversity of emotion stances includes:

[0047] For a single person, the diversity evaluation index considering the diversity of emotional stances is calculated to meet the following requirements:

[0048]

[0049] Among them, n s It is expressed as the number of different topics, l represents the list of items under consideration, i represents the i-th topic, j represents the j-th sentiment attitude, o represents the sentiment attitude (positive, negative, neutral), c(l,s i ,o j ) indicates that the list l belongs to the subject s i And the attitude is o j The number of items.

[0050] In a specific application example, the embodiment of this application assumes that there are a total of 2 topics. System A recommends 10 items, and the number of items covering each topic and each emotional stance is (5, 1, 1) and (3, 0, 0) respectively. System B also recommends 10 items, and the number of items covering each topic and each emotional stance is (3, 2, 2) and (1, 1, 1) respectively. Then the SDIndex of system A is greater than the SD Index of system B:

[0051] Based on this, the embodiment of the present application believes that the emotional stance diversity of system B is better than that of system A for determining individual-level diversity.

[0052] In some embodiments, based on the emotion stance detection results, calculating a diversity evaluation index that considers the diversity of emotion stances includes:

[0053] For several people, the diversity evaluation index considering the diversity of emotional stances satisfies:

[0054]

[0055] Among them, n s It is expressed as the number of different topics, k represents the kth sentiment attitude, j represents the jth sentiment attitude, e i,j It is expressed as the number of occurrences of items with item topic i and sentiment stance j, e i,k It is expressed as the number of occurrences of items with project topic i and sentiment stance k. A smaller Gini index in the embodiment of the present application indicates that the distribution of occurrences of items with different sentiment stances in the output results is more fair, indicating higher diversity, because different items have more opportunities to be recommended.

[0056] The embodiment of the present application proposes a recommendation system diversity detection method that takes into account the diversity of emotional stance by incorporating the diversity of the emotional stance dimension into the recommendation system diversity index. The embodiment method of the present application provides a diversity evaluation index that takes into account the emotional stance from two aspects. It makes up for the shortcomings of the indicator considerations in the existing technology. When designing the recommendation algorithm training, if this indicator is adopted, it can improve the diversity performance and reduce the current "information cocoon" and "emotional polarization" phenomena. It also provides a corresponding detection method, forming a complete recommendation algorithm detection framework and method.

[0057] An embodiment of the present application further proposes a device for detecting the diversity of recommendation results of a recommendation system, comprising a processor and a memory, wherein a computer program is stored on the memory, and when the computer program is executed by the processor, the steps of the method for detecting the diversity of recommendation results of a recommendation system as described above are implemented.

[0058] An embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the aforementioned method for detecting the diversity of recommendation results of a recommendation system are implemented.

[0059] It should be noted that, in the various embodiments of the present application, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0060] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0061] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0062] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of this application, ordinary technicians in this field can also make many forms without departing from the purpose of this application and the scope of protection of the claims, all of which are protected by this application.

Claims

1. A method for detecting the diversity of recommendation results of a recommendation system, characterized in that: include: Obtaining recommendation results from the recommendation system, and classifying the recommendation results based on a preset topic classification model to obtain topic categories of recommended content; Determine similar topics in the recommended content category and perform sentiment detection using a preset sentiment detection model. Based on the results of the emotional stance detection, a diversity evaluation index that considers the diversity of emotional stances is calculated; Based on the results of sentiment stance detection, the diversity evaluation indicators that consider the diversity of sentiment stances are calculated, including: For a single person, the diversity evaluation index that considers the diversity of emotional stances is calculated as: in, Expressed as the number of different topics, represents a list of projects under consideration, Indicates the Themes, Indicates the emotional stance and attitude, o represents emotional stance and attitude, c(l, express List of topics And the attitude is the number of projects; Based on the results of sentiment stance detection, the diversity evaluation indicators that consider the diversity of sentiment stances are calculated, including: For several people, the diversity evaluation index considering the diversity of emotional positions is calculated as: in, It is expressed as the number of different topics, k represents the kth sentiment attitude, j represents the jth sentiment attitude, , It is expressed as the number of occurrences of items with item topic i and sentiment stance j, It is expressed as the number of occurrences of items whose topic is i and whose sentiment stance is k.

2. The method for detecting the diversity of recommendation results of a recommendation system according to claim 1, wherein: Obtaining the recommendation results of the recommendation system includes: For a single person, obtain personal user information and personal recommendation results; For multiple people, multiple user information and recommendation results of the recommendation system for multiple people are obtained, and the multiple recommendation results are aggregated.

3. The method for detecting the diversity of recommendation results of a recommendation system according to claim 1, wherein: Emotional stance detection using a preset emotional stance detection model includes: Emotional stance detection is performed using a preset emotional stance detection model to classify emotional stances into positive, neutral, and negative.

4. A device for detecting the diversity of recommendation results of a recommendation system, characterized in that: The system comprises a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the steps of the method for detecting the diversity of recommendation results of the recommendation system according to any one of claims 1 to 3 are implemented.

5. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for detecting the diversity of recommendation results of a recommendation system according to any one of claims 1 to 3.

Images