Information cocoon control method and device

Abstract

The application provides an information cocoon house control method and device, and relates to the technical field of information dissemination control. The method comprises the following steps: obtaining the number of positive feedbacks of a user, the number of negative feedbacks of the user, the total number of recommendations of the user and the free exploration intensity of the user, calculating the ratio between the number of negative feedbacks of the user and the number of positive feedbacks of the user, and taking the similarity calculation result between the number of positive feedbacks of the user and the total number of recommendations of the user as the user similarity recommendation intensity; determining the recommendation system state according to the ratio, the user similarity recommendation intensity and the position of the free exploration intensity of the user in a pre-acquired information cocoon house phase diagram; determining an information cocoon house control strategy corresponding to the recommendation system state, and controlling the recommendation system according to the information cocoon house control strategy. The device executes the above method. The method and device provided in the application embodiment can reveal the information cocoon house formation mechanism, and fundamentally solve the information cocoon house problem.

Description

Technical Field

[0001] This invention relates to the field of information dissemination and control technology, specifically to an information cocoon control method and device. Background Technology

[0002] With the rapid development of information technology, people face a serious problem of information overload, namely the dilemma and challenge individuals encounter when acquiring, processing, and understanding large amounts of information. To address this problem, recommender systems have emerged. By analyzing users' interests, behaviors, and preferences, recommender systems provide personalized and precise recommendations, thereby helping users discover information, products, or services that match their interests.

[0003] Currently, recommender systems are widely used in e-commerce, social media, video streaming, music, and news. Traditional recommender systems are mainly based on collaborative filtering, content filtering, and hybrid filtering methods. Collaborative filtering methods rely on users' historical behavior and interests, finding other users with similar interests and recommending content they like to the target user. Content filtering methods rely on content analysis and tagging to recommend content related to the user's interests. Hybrid filtering methods combine collaborative filtering and content filtering to provide more comprehensive and accurate recommendation results.

[0004] However, current recommender systems still have some problems. Among them, the most worrying is the formation of information cocoons. Due to the filtering effect of recommender systems, users are confined to single, homogeneous content, unable to freely access a wide range of diverse information. Information cocoons have significant negative impacts: on the one hand, they damage user experience; on the other hand, they impair individual cognition, causing users to become addicted to personal satisfaction and become completely indifferent to or neglectful of other information, consciously or unconsciously isolating themselves in a narrow circle. To address the problem of recommendation homogeneity, existing research has proposed a series of diverse recommendation methods, such as post-processing, determinant-based processes, and ranking learning. However, these methods only diversify recommendation results to a certain extent and do not reveal the formation mechanism of information cocoons, failing to fundamentally solve the problem. Summary of the Invention

[0005] To address the problems in the prior art, embodiments of the present invention provide an information cocoon control method and apparatus, which can at least partially solve the problems existing in the prior art.

[0006] On the one hand, this invention proposes a method for controlling information cocoons, comprising:

[0007] The system obtains the number of positive feedback users, the number of negative feedback users, the total number of recommendations, and the intensity of free exploration of the recommended users in the recommendation system to be subject to information cocoon control. It also calculates the ratio between the number of negative feedback users and the number of positive feedback users, and uses the similarity calculation result between the number of positive feedback users and the total number of recommendations users as the recommendation intensity based on the similarity of the recommended users.

[0008] The state of the recommendation system is determined based on the positions of the ratio, the recommendation strength based on the similarity of the recommended users, and the free exploration strength of the recommended users in the pre-acquired information cocoon phase transition diagram.

[0009] Determine the information cocoon control strategy corresponding to the state of the recommendation system, and perform information cocoon control on the recommendation system according to the information cocoon control strategy.

[0010] The step of determining the recommendation system state based on the positions of the ratio, the recommendation strength based on the similarity of the recommended users, and the free exploration strength of the recommended users in the pre-acquired information cocoon phase transition diagram includes:

[0011] If the location is determined to be within a first spatial region corresponding to diversification, then the recommendation system state is determined to be a diversification state.

[0012] If the location is determined to be in the second spatial region corresponding to a mild information cocoon, then the recommendation system state is determined to be a mild information cocoon state.

[0013] If the location is determined to be in a third spatial region corresponding to the deep information cocoon, then the recommendation system state is determined to be the deep information cocoon state.

[0014] The step of determining the information cocoon control strategy corresponding to the state of the recommendation system includes:

[0015] If the recommendation system is determined to be in the state of deep information cocoon, and the ratio is lower than a preset ratio threshold, and the intensity of free exploration by recommended users is lower than a preset free exploration intensity threshold, then the information cocoon control strategy is determined to be to increase the utilization rate of negative samples and enhance the intensity of free exploration by users.

[0016] If the recommendation system is determined to be in the state of the deep information cocoon, and the recommendation strength based on user similarity is higher than a preset similarity recommendation strength threshold, while the recommendation strength based on user free exploration is lower than a preset free exploration strength threshold, then the information cocoon control strategy is determined to be to suppress the use of similarity recommendations and enhance the intensity of user free exploration.

[0017] The acquisition of the information cocoon phase transition diagram includes:

[0018] Obtaining the information cocoon phase transition diagram includes:

[0019] Data measurement results are obtained based on user interaction data; the data measurement results include the correlation strength between the recommendation distribution information entropy and the positive feedback ratio of observed users, the negative feedback ratio of observed users, the recommendation strength of observed users similarity, and the intensity of observed users' free exploration, as well as the global popularity of different item types and the correlation matrix between different item types;

[0020] Based on the observation preference vector and correlation matrix of the observed users determined by the global popularity, the similarity between the observation preference vector and the item feature vector is calculated. Based on the similarity between the observation preference vector and the item feature vector and the recommendation strength of the observation user similarity, the recommendation probability of recommending each item to the observed user is determined, and each item is recommended to the observed user with the recommendation probability to obtain a dataset.

[0021] The similarity between the observed user's inherent preference vector and the item feature vector, determined based on the global popularity, is used as the probability of accepting the recommendation. Positive feedback items with a probability greater than the probability of accepting the recommendation are selected from the dataset as positive feedback samples, and the remaining items are used as negative feedback samples.

[0022] A stochastic differential equation is constructed based on the proportion of positive feedback from observed users, the proportion of negative feedback from observed users, positive feedback samples and negative feedback samples, the interaction function between the observed preference vector and the item feature vector, the intensity of free exploration by observed users, and the standard Wiener process, and the observed preference vector is updated based on the stochastic differential equation.

[0023] The Falk-Planck equation is obtained from the stochastic differential equation, and the derivative of the Falk-Planck equation with respect to time is obtained to obtain the recommendation information entropy distribution.

[0024] The relative information entropy distribution is calculated based on the recommended information entropy distribution and the inherent information entropy distribution.

[0025] Based on the relative information entropy distribution, a first spatial region corresponding to diversification, a second spatial region corresponding to light information cocoons, and a third spatial region corresponding to deep information cocoons are determined. Relevant parameters of the information cocoon phase transition diagram are obtained through simulation. The information cocoon phase transition diagram is drawn based on the relevant parameters, the first spatial region, the second spatial region, and the third spatial region.

[0026] The step of obtaining data measurement results based on user interaction data includes:

[0027] A dataset is constructed based on recommendation logs and feedback logs from active users; the dataset includes item types corresponding to the items.

[0028] From the dataset, we obtain behavioral features that indicate liking for recommended items to obtain positive feedback behavior; from the dataset, we obtain behavioral features that indicate disliking for recommended items to obtain negative feedback behavior.

[0029] The ratio of positive feedback behavior to recommendation behavior is used as the positive feedback ratio of observed users, and the ratio of negative feedback behavior to recommendation behavior is used as the negative feedback ratio of observed users; the number of recommendation behavior is the sum of the number of positive feedback behavior and the number of negative feedback behavior.

[0030] Establish the first correlation strength between the recommendation information entropy distribution and the proportion of positive feedback from observed users, and establish the second correlation strength between the recommendation information entropy distribution and the proportion of negative feedback from observed users.

[0031] The step of obtaining data measurement results based on user interaction data includes:

[0032] The average value of the positive feedback behavior distribution is used as the global popularity of different item types;

[0033] Establish a third correlation strength between the recommendation information entropy distribution and global popularity.

[0034] The step of obtaining data measurement results based on user interaction data includes:

[0035] The covariance of the positive feedback behavior distribution is used as the correlation matrix between different item types;

[0036] Establish a fourth correlation strength between the entropy distribution of recommendation information and the correlation matrix.

[0037] On one hand, the present invention proposes an information cocoon control device, comprising:

[0038] The acquisition unit is used to acquire the number of positive feedback users, the number of negative feedback users, the total number of recommendations, and the intensity of free exploration of the recommended users in the recommendation system to be controlled by information cocoon, and to calculate the ratio between the number of negative feedback users and the number of positive feedback users, and to use the similarity calculation result between the number of positive feedback users and the total number of recommendations as the recommendation intensity of the similarity of the recommended users.

[0039] The determining unit is used to determine the state of the recommendation system based on the ratio, the recommendation strength of the similarity between the recommended users, and the position of the free exploration strength of the recommended users in the pre-acquired information cocoon phase transition diagram;

[0040] The control unit is used to determine the information cocoon control strategy corresponding to the state of the recommendation system, and to perform information cocoon control on the recommendation system according to the information cocoon control strategy.

[0041] In another aspect, embodiments of the present invention provide an electronic device, including: a processor, a memory, and a bus, wherein,

[0042] The processor and the memory communicate with each other via the bus;

[0043] The memory stores program instructions that can be executed by the processor, and the processor can execute the following methods by calling the program instructions:

[0044] The system obtains the number of positive feedback users, the number of negative feedback users, the total number of recommendations, and the intensity of free exploration of the recommended users in the recommendation system to be subject to information cocoon control. It also calculates the ratio between the number of negative feedback users and the number of positive feedback users, and uses the similarity calculation result between the number of positive feedback users and the total number of recommendations users as the recommendation intensity based on the similarity of the recommended users.

[0045] The state of the recommendation system is determined based on the positions of the ratio, the recommendation strength based on the similarity of the recommended users, and the free exploration strength of the recommended users in the pre-acquired information cocoon phase transition diagram.

[0046] Determine the information cocoon control strategy corresponding to the state of the recommendation system, and perform information cocoon control on the recommendation system according to the information cocoon control strategy.

[0047] This invention provides a non-transitory computer-readable storage medium, comprising:

[0048] The non-transitory computer-readable storage medium stores computer instructions that cause the computer to perform the following methods:

[0049] The system obtains the number of positive feedback users, the number of negative feedback users, the total number of recommendations, and the intensity of free exploration of the recommended users in the recommendation system to be subject to information cocoon control. It also calculates the ratio between the number of negative feedback users and the number of positive feedback users, and uses the similarity calculation result between the number of positive feedback users and the total number of recommendations users as the recommendation intensity based on the similarity of the recommended users.

[0050] The state of the recommendation system is determined based on the positions of the ratio, the recommendation strength based on the similarity of the recommended users, and the free exploration strength of the recommended users in the pre-acquired information cocoon phase transition diagram.

[0051] Determine the information cocoon control strategy corresponding to the state of the recommendation system, and perform information cocoon control on the recommendation system according to the information cocoon control strategy.

[0052] The information cocoon control method and apparatus provided in this invention acquire the number of positive feedback users, the number of negative feedback users, the total number of recommendations, and the intensity of free exploration of recommending users in a recommendation system to be controlled for information cocoons. It calculates the ratio between the number of negative feedback users and the number of positive feedback users, and uses the similarity calculation result between the number of positive feedback users and the total number of recommendations as the recommendation intensity based on the similarity. Based on the positions of the ratio, the recommendation intensity, and the intensity of free exploration of recommending users in a pre-acquired information cocoon phase transition diagram, it determines the state of the recommendation system. It then determines the information cocoon control strategy corresponding to the state of the recommendation system and performs information cocoon control on the recommendation system according to the information cocoon control strategy. This reveals the formation mechanism of information cocoons and fundamentally solves the information cocoon problem. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0054] Figure 1 This is a flowchart illustrating an embodiment of the information cocoon control method provided by the present invention.

[0055] Figure 2 This is a schematic diagram illustrating the phase transition diagram of the information cocoon provided in an embodiment of the present invention.

[0056] Figure 3 This is a schematic diagram illustrating the modularity of the information cocoon control method provided in the embodiments of the present invention.

[0057] Figure 4 This is a flowchart illustrating an information cocoon control method provided in another embodiment of the present invention.

[0058] Figure 5 This is a schematic diagram of the structure of an information cocoon control device provided in an embodiment of the present invention.

[0059] Figure 6 This is a schematic diagram of the physical structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Here, the illustrative embodiments and descriptions of the present invention are used to explain the present invention, but are not intended to limit the present invention. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be arbitrarily combined with each other.

[0061] Figure 1 This is a flowchart illustrating an embodiment of the information cocoon control method provided by the present invention, as shown below. Figure 1 As shown, the information cocoon control method provided in this embodiment of the invention includes:

[0062] Step S1: Obtain the number of positive feedback users, the number of negative feedback users, the total number of recommendations, and the intensity of free exploration of the recommended users in the recommendation system to be subject to information cocoon control. Calculate the ratio between the number of negative feedback users and the number of positive feedback users, and use the similarity calculation result between the number of positive feedback users and the total number of recommendations users as the recommendation intensity based on user similarity.

[0063] Step S2: Determine the state of the recommendation system based on the position of the ratio, the recommendation strength based on the similarity of the recommended users, and the free exploration strength of the recommended users in the pre-acquired information cocoon phase transition diagram.

[0064] Step S3: Determine the information cocoon control strategy corresponding to the state of the recommendation system, and perform information cocoon control on the recommendation system according to the information cocoon control strategy.

[0065] In step S1 above, the device acquires the number of positive feedback users, the number of negative feedback users, the total number of recommendations, and the intensity of free exploration by the recommending users in the recommendation system to be controlled by information cocoons. It then calculates the ratio between the number of negative feedback users and the number of positive feedback users, and uses the similarity calculation result between the number of positive feedback users and the total number of recommendations as the recommendation strength based on user similarity. The device can be a computer device that performs this method, such as a server.

[0066] In step S2 above, the device determines the state of the recommendation system based on the positions of the ratio, the recommendation strength based on the similarity of the recommended users, and the free exploration strength of the recommended users in the pre-acquired information cocoon phase transition diagram. For example... Figure 2 The diagram shown is a pre-acquired information cocoon phase transition diagram, where the length direction corresponds to the recommendation strength based on the similarity between recommended users, the height direction corresponds to the aforementioned ratio, and the depth direction corresponds to the recommendation strength based on the freedom of users to explore.

[0067] The step of determining the state of the recommendation system based on the positions of the ratio, the recommendation strength based on the similarity of the recommended users, and the free exploration strength of the recommended users in the pre-acquired information cocoon phase transition diagram includes:

[0068] If the location is determined to be within a first spatial region corresponding to diversification, then the recommendation system state is determined to be a diversification state; the first spatial region corresponds to Figure 2 The dark area at the top shows that when the system is in a diversified state, the distribution of relative information entropy is a unimodal distribution with a mean greater than 1. This indicates that the information diversity provided by the current recommendation system exceeds the user's own demand for diversity.

[0069] If the location is determined to be within the second spatial region corresponding to a mild information cocoon, then the recommendation system state is determined to be a mild information cocoon state; the second spatial region corresponds to Figure 2 In the light-colored area in the middle, when the system is in a state of mild information cocoon, the distribution of relative information entropy is a unique bimodal distribution. The diverse preferences of some users are well satisfied, but some users are still confined to monotonous information.

[0070] If the location is determined to be within a third spatial region corresponding to the deep information cocoon, then the recommendation system state is determined to be a deep information cocoon state. The third spatial region corresponds to... Figure 2 In the darkest area at the bottom, when the system is in a deep information cocoon state, the distribution of relative information entropy returns to a unimodal distribution with a mean of less than 0.5. This indicates that the diversity of users' intrinsic preferences is greatly suppressed by the recommendation system, trapping them in a deep information cocoon.

[0071] The parameter range of the current system can be determined based on the shape of the relative information entropy distribution in the current real data. Positive sample rate, negative sample rate, and similarity-based recommendation strength can further determine the current system parameters and state. It is worth noting that the recommendation information entropy distribution has similarly undergone a phase transition process from a "unimodal-bimodal-unimodal" distribution.

[0072] In step S3 above, the device determines an information cocoon control strategy corresponding to the state of the recommendation system, and performs information cocoon control on the recommendation system according to the information cocoon control strategy.

[0073] The determination of the information cocoon control strategy corresponding to the state of the recommendation system includes:

[0074] If the recommendation system is determined to be in the deep information cocoon state, and the ratio is lower than a preset ratio threshold, and the intensity of free exploration by recommended users is lower than a preset free exploration intensity threshold, then the information cocoon control strategy is determined to be to increase the utilization rate of negative samples and enhance the intensity of free exploration by users; Figure 2 As shown, when the system operates at the parameter represented by the pentagram in the lower right corner, it is trapped in a deep information cocoon. This information cocoon arises from excessively low relative utilization of negative samples and insufficient free exploration, rather than excessively strong similarity-based recommendation. Therefore, for a system with this parameter, the optimal solution to escape the information cocoon is to increase the utilization rate of negative samples and enhance the intensity of user free exploration.

[0075] If the recommendation system is determined to be in the deep information cocoon state, and the recommendation strength based on user similarity is higher than a preset similarity recommendation strength threshold, while the user's free exploration strength is lower than a preset free exploration strength threshold, then the information cocoon control strategy is determined to be to suppress the use of similarity recommendations and enhance the user's free exploration strength. Figure 2 As shown, when the system operates at the parameters represented by the pentagram in the upper left corner, it is also trapped in a deep information cocoon, but for different reasons: this system suffers from excessively strong similarity-based recommendation and insufficient free exploration. Therefore, to address the information cocoon problem in this system, it is necessary to curb the abuse of similarity-based recommendation methods and encourage users' free exploration behavior.

[0076] If the recommendation system is determined to be in a state of mild information cocoon and diversity, no processing is required.

[0077] Obtaining the information cocoon phase transition diagram includes:

[0078] Data measurement results are obtained based on user interaction data. These results include the correlation strength between the recommendation information entropy distribution and the proportion of positive feedback from observed users, the proportion of negative feedback from observed users, the similarity recommendation strength of observed users, and the intensity of free exploration by observed users, as well as the global popularity of different item types and the correlation matrix between different item types. It should be noted that observed users refer to users in the information cocoon phase transition diagram acquisition stage, while recommended users refer to users in the information cocoon phase transition diagram usage stage. The content corresponding to observed users (such as the intensity of free exploration by observed users) and the content corresponding to recommended users (such as the intensity of free exploration by recommended users) have the same meaning.

[0079] The step of obtaining data measurement results based on user interaction data includes:

[0080] A dataset is constructed based on recommendation and feedback logs of active users; the dataset includes item types corresponding to the items; active users are those who have interacted with the recommendation system more than a certain threshold, selected from the interaction logs of newly registered users on the platform, to ensure that the users in the dataset meet the following two conditions:

[0081] (1) The entire interaction history on the platform.

[0082] (2) There is sufficient interaction with the recommendation system on this platform.

[0083] Behavioral features representing liking recommended items are extracted from the dataset to obtain positive feedback behaviors, and behavioral features representing disliking recommended items are extracted from the dataset to obtain negative feedback behaviors. Positive feedback behaviors include, for example, liking and completing a video. Negative feedback behaviors include, for example, disliking and ignoring. Since explicit negative feedback behaviors, such as disliking, are very sparse, implicit negative feedback behaviors, i.e., ignoring, are used here for subsequent measurement.

[0084] The ratio of positive feedback behavior to recommendation behavior is used as the positive feedback ratio of observed users, and the ratio of negative feedback behavior to recommendation behavior is used as the negative feedback ratio of observed users; the number of recommendation behavior is the sum of the number of positive feedback behavior and the number of negative feedback behavior.

[0085] We establish the first correlation strength between the recommendation information entropy distribution and the proportion of positive feedback from observed users, and the second correlation strength between the recommendation information entropy distribution and the proportion of negative feedback from observed users. This allows us to obtain the frequency distribution of the total number of recommendations received by observed user l across different item types. Here, N represents the number of types. Similarly, we can obtain the frequency distribution of positive feedback numbers across different item types. Through calculation and By calculating the similarity between the recommendations, we can obtain the strength parameter of the similarity-based recommendations. This is achieved by calculating the information entropy of the recommendation distribution, i.e., the recommendation information entropy distribution. This allows for a preliminary measurement of information cocoons. Next, we establish the first correlation strength between the distribution of recommended information entropy and the proportion of positive feedback from observed users, and the second correlation strength between the distribution of recommended information entropy and the proportion of negative feedback from observed users.

[0086] The step of obtaining data measurement results based on user interaction data includes:

[0087] The average value of the positive feedback behavior distribution is used as the global popularity of different item types; the sample data for calculating the average value of the positive feedback behavior distribution can be selected from a population that includes as many observed users as possible.

[0088] Establish a third correlation strength between the recommendation information entropy distribution and global popularity.

[0089] The step of obtaining data measurement results based on user interaction data includes:

[0090] The covariance of the positive feedback behavior distribution is used as the correlation matrix between different item types; the sample data for calculating the covariance of the positive feedback behavior distribution can be selected from a population that includes as many observed users as possible.

[0091] Establish a fourth correlation strength between the entropy distribution of recommendation information and the correlation matrix.

[0092] The step of obtaining data measurement results based on user interaction data includes:

[0093] Establish the fifth correlation strength between the recommendation information entropy distribution and the recommendation strength based on the similarity between observed users, and establish the sixth correlation strength between the recommendation information entropy distribution and the intensity of free exploration by observed users.

[0094] Based on the observed user's observation preference vector and correlation matrix determined by the global popularity, the similarity between the observation preference vector and the item feature vector is calculated. Based on the similarity between the observation preference vector and the item feature vector, and the recommendation strength based on the observed user similarity, the recommendation probability of recommending each item to the observed user is determined, and each item is recommended to the observed user according to the recommendation probability, thus obtaining a dataset. The observed user's observation preference vector is determined based on the global popularity. include:

[0095] Simulations were conducted based on a minimal implementation model, in which each observed user l has an inherent preference distribution across different N item types. and an observed preference distribution

[0096] These vectors are initialized from two empirical datasets. Following existing work, it is assumed that the intrinsic preference distribution of observed users follows a Dirichlet distribution. Where μ user This is a global popularity vector obtained from empirical data, corresponding to the global popularity mentioned above. Furthermore, to avoid the observed preferences deviating too far from the global popularity, the same Dirichlet distribution is used. This is used to initialize the observation of user preferences. Each item k has an item feature vector. This is a one-hot vector used to encode the topic to which an item belongs. Candidate items are randomly selected from the item pool in each dataset.

[0097] The similarity between the observation preference vector and the item feature vector is calculated using the following expression.

[0098]

[0099] in,

[0100]

[0101] Φ is the correlation matrix.

[0102] Similarity can be measured using metrics such as inner product, cosine similarity, and Jensen-Shannon divergence.

[0103] The probability p of recommending each item to the observed user is calculated using the following expression. lk :

[0104]

[0105] Where β represents the recommendation strength based on observed user similarity, β = 0 represents a purely random recommendation strategy, and the larger the β, the more likely items with higher similarity scores are to be recommended. I is the set of all candidate items, l k' These are candidate items. The dataset is denoted as R. β .

[0106] The similarity between the observed user's inherent preference vector and the item feature vector, determined based on the global popularity, is used as the probability of accepting the recommendation. Items with a probability greater than the probability of accepting the recommendation are selected from the dataset as positive feedback samples, and the remaining items are used as negative feedback samples. The observed user's inherent preference vector is determined based on global popularity. Please refer to the above explanation. The probability of accepting a recommendation can be calculated using the following expression.

[0107]

[0108] Positive feedback samples are denoted as Negative feedback samples are denoted as

[0109] A stochastic differential equation is constructed based on the proportion of positive feedback from observed users, the proportion of negative feedback from observed users, positive feedback samples and negative feedback samples, the interaction function between the observed preference vector and the item feature vector, the intensity of free exploration by observed users, and the standard Wiener process, and the observed preference vector is updated based on the stochastic differential equation.

[0110] Considering that users also actively explore through other resources (such as search engines), and that these exploration behaviors are important inputs to recommendation algorithms, this invention models the active exploration behavior of users as a stochastic process, introducing σ to represent the intensity of the user's free exploration.

[0111] The expression for the stochastic differential equation is as follows:

[0112]

[0113] Where, γ + ≥0 and γ - ≤0 represent the proportion of positive feedback and the proportion of negative feedback from observed users, respectively. Physically speaking, γ+ and γ - This determines the likelihood that the recommendation system will recommend ideal and undesirable items to the observed user. This represents the observed preferences of user l. With the recommended items The interaction function between features. W t This is a standard Wiener process. It is used here. To measure and The distance between them. σ represents the intensity of the observed user's free exploration.

[0114] like Figure 3 As shown, the above technical content can be implemented modularly, including a recommendation module, a user feedback module, and an observation and update module, wherein:

[0115] The recommendation module is used to calculate the similarity between the observation preference vector and the item feature vector, and to calculate the recommendation probability of recommending each item to the observation user. It then recommends each item to the observation user based on the recommendation probability, resulting in a dataset R. β and R β Send to the user feedback module.

[0116] The user feedback module is used to calculate the probability of accepting a recommendation, and to determine the probability based on the observed user's free exploration intensity, the probability of accepting a recommendation, and R. β Wait, and obtain positive feedback samples. and negative feedback samples And and Send to the observation update module.

[0117] The observation update module is used to update the observation based on... and We construct stochastic differential equations and update the observation preference vectors based on these equations.

[0118] The Falk-Planck equation is obtained from the stochastic differential equation, and the derivative of the Falk-Planck equation with respect to time is taken to obtain the recommendation information entropy distribution; the Falk-Planck equation is derived theoretically as follows:

[0119]

[0120] in, The calculation represents the observed preferences of user l at time t, discovered through a recommendation algorithm, in a specific item category j. The probability. Note that the first term on the right-hand side of the formula captures both similarity-based recommendations (β) and positive / negative feedback (γ). ±The first term describes the effect of the recommendation algorithm under the first mechanism, while the second term describes the self-exploration (σ) of the observed user. This equation shows that the adaptive information dynamics in the human-computer interaction system are jointly driven by recommendation and self-exploration. Considering that the system reaches a static state, the mean field approximation is used here. After derivation, the recommendation information entropy distribution can be obtained as:

[0121] P(s)=P s =P(s) (1) )*P(s (2) )*…*P(s (j) )*…*P(s (N) )

[0122] Among them, the information entropy distribution of recommendations

[0123] The relative information entropy distribution is calculated based on the recommended information entropy distribution and the inherent information entropy distribution; for example... Figure 4 As shown, by constructing stochastic differential equations, an adaptive information dynamics model is built. This invention enables a more rigorous definition of the information cocoon, which is the ratio between the information entropy of the recommended items observed by the user and the information entropy of the user's intrinsic preferences, i.e., relative information entropy.

[0124] s represents the distribution of recommendation information entropy. Indicates user's intrinsic preferences The entropy, i.e., the inherent information entropy distribution, is as follows, referring to the above explanation: The relative information entropy distribution is as follows:

[0125]

[0126] Wherein, P(s) * ) is an arbitrary probability density function, representing the inherent information entropy distribution of the observed user.

[0127] Based on the relative information entropy distribution, a first spatial region corresponding to diversification, a second spatial region corresponding to mild information cocoons, and a third spatial region corresponding to deep information cocoons are determined. Relevant parameters of the information cocoon phase transition diagram are obtained through simulation. The information cocoon phase transition diagram is then drawn based on these parameters, the first spatial region, the second spatial region, and the third spatial region. Furthermore, based on the above theoretical derivation, three phases of the information cocoon can be defined:

[0128] (diversification)

[0129] (Mild information cocoon)

[0130] (Deep Information Cocoon)

[0131] Through large-scale simulations, four parameters can be determined: the strength of the recommendation based on observed user similarity (β), the proportion of positive feedback from observed users (γ), and so on. + ), observe the proportion of negative user feedback (γ) - The similarity between observers and the intensity of free exploration by observers (σ) determine the phase of the system. Higher observer-user similarity recommendation intensity and a higher proportion of positive feedback from observers make the system more biased towards the deep information cocoon phase; however, higher proportions of positive feedback from observers and a higher intensity of free exploration by observers make the system more biased towards the diversified phase. A complete information cocoon phase transition diagram can be drawn through theoretical analysis and simulation experiments under different parameters.

[0132] Combined with specific application scenarios Figure 4 Further explanation:

[0133] Cleaning and measurement of user interaction data:

[0134] Taking short video recommendation and news recommendation scenarios as examples, this paper explains in detail how to clean and measure user interaction data. Short video recommendation and news recommendation scenarios are currently the two most severely affected by information cocoons. Research on these two scenarios can help to better determine the formation mechanism and control methods of information cocoons.

[0135] Specifically, in the short video scenario, short videos involve 20 different video categories. Positive feedback behavior refers to user likes and views, while negative feedback refers to exposure but neglect. By analyzing the distribution of user viewing time across different categories, the global popularity of each video category can be obtained. By calculating the covariance of the user viewing time distribution, the correlation matrix of each video category can be obtained.

[0136] In the news context, the news encompasses 14 different categories, such as weather and sports. Positive feedback behavior refers to user clicks, while negative feedback refers to user exposure but neglect. By analyzing the distribution of user clicks across categories, the global popularity of each news category can be obtained. By calculating the covariance of the distribution of user viewing time, the correlation matrix of each news category can be obtained.

[0137] Construction of the adaptive information dynamics model and characterization of phase transition diagrams based on the adaptive information dynamics model:

[0138] Based on the aforementioned short video and news scenarios, a minimal implementation of the generalized model was proposed. Specifically, a linear model was used. To explain the observed preferences and item characteristics Distance in the embedding space. Besides a linear function, there are many other choices for the mathematical form of this function, such as... However, when the higher-order terms are relatively weak, all candidate functions can be expanded into Taylor series. Furthermore, it is assumed that the probability of user l accepting the recommended item k is... Measurement and The similarity between them is calculated using the formula: Here, the inner product can be used as the similarity function, or other functions can be used instead.

[0139] Furthermore, simulations were conducted based on a minimal implementation model. In this minimal implementation, each user l has an inherent preference distribution across different N item types. and an observed preference distribution These vectors are initialized from two empirical datasets. Based on existing work, it is assumed that the user's intrinsic preference distribution follows a Dirichlet distribution. Where μ user This is a global popularity vector for each category, obtained from empirical data. Furthermore, to avoid observed preferences deviating too far from the global popularity, the same Dirichlet distribution is used. This is used to initialize the user's observed preferences. Each item k has a fixed feature vector. This is a one-hot vector used to encode the topic to which an item belongs. Candidate items are randomly selected from the item pool in each dataset. The specific process is as follows.

[0140] At each time step t, the interaction between the user and the recommendation algorithm is repeated in the following manner:

[0141] (1) The recommendation algorithm recommends a set of items R to each user l. β N rec There are 10 different items. The probability of each item being recommended follows the recommendation formula described above.

[0142] (2) The probability that user l gives positive feedback to recommended item k is: The probability of giving negative feedback is Items that receive positive feedback form a set of positive feedback samples. Other items form a group of negative feedback samples.

[0143] (3) User l performs random self-exploration according to the Wiener process.

[0144] (4) The recommendation algorithm updates the observed preferences of each user l according to the above stochastic differential equation.

[0145] State determination and control strategy generation based on information cocoon phase transition diagram:

[0146] Based on the above empirical analysis, theoretical analysis, and simulation, phase transition diagrams for short video and news scenarios can be obtained. Considering the antagonistic effect of positive and negative feedback on information cocoon emergence, the relative utilization rate of feedback is defined here.

[0147] |γ - / γ + | represents the update rate of negative feedback relative to positive feedback. A three-dimensional state diagram of video and news recommendations can be displayed, revealing two phase transitions: from diversification to mild information cocoons, and then from mild information cocoons to deep information cocoons. The phase transitions occurring under critical parameters indicate that the generation of information cocoons can be controlled by appropriately balancing the aforementioned factors. Indeed, even in the current system where similarity-based strength β dominates, a slightly increased self-exploration rate σ and the relative utilization rate of negative feedback |γ... - / γ + This will keep the system away from information cocoons. This indicates that a balance between positive and negative feedback, and between similarity-based recommendations and initiatives that encourage self-exploration, effectively inhibits the formation of information cocoons.

[0148] The information cocoon control method provided in this invention obtains the number of positive feedback users, the number of negative feedback users, the total number of recommendations, and the intensity of free exploration by the recommending users in the recommendation system to be controlled. It calculates the ratio between the number of negative feedback users and the number of positive feedback users, and uses the similarity calculation result between the number of positive feedback users and the total number of recommendations as the recommendation intensity based on user similarity. The method determines the state of the recommendation system based on the positions of the ratio, the recommendation intensity, and the intensity of free exploration by the recommending users in a pre-acquired information cocoon phase transition diagram. It then determines the information cocoon control strategy corresponding to the state of the recommendation system and performs information cocoon control on the recommendation system according to the strategy. This method can reveal the formation mechanism of information cocoons and fundamentally solve the information cocoon problem.

[0149] Further, determining the state of the recommendation system based on the positions of the ratio, the recommendation strength based on the similarity of the recommended users, and the free exploration strength of the recommended users in the pre-acquired information cocoon phase transition diagram includes:

[0150] If the location is determined to be within a first spatial region corresponding to diversification, then the recommendation system state is determined to be a diversification state; this can be referred to the above embodiments for explanation, and will not be repeated here.

[0151] If the location is determined to be in the second spatial region corresponding to a mild information cocoon, then the recommendation system state is determined to be a mild information cocoon state; this can be referred to the above embodiments for explanation, and will not be repeated here.

[0152] If the location is determined to be within a third spatial region corresponding to a deep information cocoon, then the recommendation system state is determined to be a deep information cocoon state. This can be referred to the above embodiments for further explanation, and will not be repeated here.

[0153] Furthermore, determining the information cocoon control strategy corresponding to the state of the recommendation system includes:

[0154] If the recommendation system is determined to be in the state of a deep information cocoon, and the ratio is lower than a preset ratio threshold, and the intensity of free exploration by recommended users is lower than a preset free exploration intensity threshold, then the information cocoon control strategy is determined to be to increase the utilization rate of negative samples and enhance the intensity of free exploration by users; this can be referred to the above embodiments for explanation, and will not be repeated here.

[0155] If the recommendation system is determined to be in the deep information cocoon state, and the recommendation strength based on user similarity is higher than a preset similarity recommendation strength threshold, while the user's free exploration strength is lower than a preset free exploration strength threshold, then the information cocoon control strategy is determined to be to suppress the use of similarity recommendations and enhance the user's free exploration strength. This can be referred to the above embodiments for explanation, and will not be repeated here.

[0156] Further, obtaining the information cocoon phase transition diagram includes:

[0157] Obtaining the information cocoon phase transition diagram includes:

[0158] Data measurement results are obtained based on user interaction data; the data measurement results include the correlation strength between the recommendation distribution information entropy and the positive feedback ratio of observed users, the negative feedback ratio of observed users, the similarity recommendation strength of observed users, and the free exploration strength of observed users, as well as the global popularity of different item types and the correlation matrix between different item types; these can be referred to the above embodiments for explanation, and will not be repeated here.

[0159] Based on the observation preference vector and correlation matrix of the observed users determined by the global popularity, the similarity between the observation preference vector and the item feature vector is calculated. Based on the similarity between the observation preference vector and the item feature vector and the recommendation strength of the observation user similarity, the recommendation probability of recommending each item to the observed user is determined, and each item is recommended to the observed user with the recommendation probability to obtain a dataset. The above embodiments can be referred to for explanation, and will not be repeated here.

[0160] The similarity between the observed user's inherent preference vector and the item feature vector, determined based on the global popularity, is used as the probability of accepting the recommendation. Items with a probability greater than the probability of accepting the recommendation are selected from the dataset as positive feedback samples, and the remaining items are used as negative feedback samples. This can be referred to the above embodiment for explanation, and will not be repeated here.

[0161] A stochastic differential equation is constructed based on the proportion of positive feedback from observed users, the proportion of negative feedback from observed users, positive feedback samples and negative feedback samples, the interaction function between the observed preference vector and the item feature vector, the intensity of free exploration by observed users, and the standard Wiener process. The observed preference vector is then updated based on the stochastic differential equation. This can be referred to the above embodiments for explanation, and will not be repeated here.

[0162] The Falk-Planck equation is obtained from the stochastic differential equation, and the derivative of the Falk-Planck equation with respect to time is obtained to obtain the recommendation information entropy distribution; the above embodiments can be referred to for explanation, and will not be repeated here.

[0163] The relative information entropy distribution is calculated based on the recommended information entropy distribution and the inherent information entropy distribution; this can be referred to the above embodiments for explanation, and will not be repeated here.

[0164] Based on the relative information entropy distribution, a first spatial region corresponding to diversification, a second spatial region corresponding to light information cocoons, and a third spatial region corresponding to deep information cocoons are determined. Relevant parameters of the information cocoon phase transition diagram are obtained through simulation. The information cocoon phase transition diagram is then drawn based on these parameters, the first spatial region, the second spatial region, and the third spatial region. This can be referred to the above embodiment for further explanation and will not be repeated here.

[0165] Furthermore, the step of obtaining data measurement results based on user interaction data includes:

[0166] A dataset is constructed based on recommendation logs and feedback logs from active users; the dataset includes item types corresponding to the items; please refer to the above embodiments for further explanation, which will not be repeated here.

[0167] The dataset is used to obtain behavioral features that indicate liking for recommended items, resulting in positive feedback behavior. Similarly, the dataset is used to obtain behavioral features that indicate disliking for recommended items, resulting in negative feedback behavior. This can be explained with reference to the above embodiments and will not be repeated here.

[0168] The ratio of positive feedback behavior to recommendation behavior is used as the positive feedback ratio of observed users, and the ratio of negative feedback behavior to recommendation behavior is used as the negative feedback ratio of observed users; the number of recommendation behavior is the sum of the number of positive feedback behavior and the number of negative feedback behavior; the above embodiments can be referred to for explanation, and will not be repeated here.

[0169] Establish a first correlation strength between the recommendation information entropy distribution and the proportion of positive feedback from observed users, and establish a second correlation strength between the recommendation information entropy distribution and the proportion of negative feedback from observed users. Refer to the above embodiments for further explanation; details will not be repeated here.

[0170] Furthermore, the step of obtaining data measurement results based on user interaction data includes:

[0171] The average value of the positive feedback behavior distribution is used as the global popularity of different item types; this can be referred to the above embodiments for explanation, and will not be repeated here.

[0172] A third correlation strength is established between the recommendation information entropy distribution and global popularity. This can be explained with reference to the above embodiments and will not be repeated here.

[0173] Furthermore, the step of obtaining data measurement results based on user interaction data includes:

[0174] The covariance of the positive feedback behavior distribution is used as the correlation matrix between different item types; this can be referred to the above embodiments for explanation, and will not be repeated here.

[0175] A fourth correlation strength is established between the recommendation information entropy distribution and the correlation matrix. This can be explained with reference to the above embodiments and will not be repeated here.

[0176] Figure 5 This is a schematic diagram of the structure of an information cocoon control device provided in an embodiment of the present invention, as shown below. Figure 5 As shown, the information cocoon control device provided in this embodiment of the invention includes an acquisition unit 501, a determination unit 502, and a control unit 503, wherein:

[0177] The acquisition unit 501 is used to acquire the number of positive feedback users, the number of negative feedback users, the total number of recommendations, and the intensity of free exploration of the recommended users in the recommendation system to be subject to information cocoon control, and to calculate the ratio between the number of negative feedback users and the number of positive feedback users, and to use the similarity calculation result between the number of positive feedback users and the total number of recommendations users as the recommendation intensity based on the similarity between the two users; the determination unit 502 is used to determine the state of the recommendation system based on the position of the ratio, the recommendation intensity based on the similarity between the two users, and the intensity of free exploration of the two users in the pre-acquired information cocoon phase transition diagram; the control unit 503 is used to determine the information cocoon control strategy corresponding to the state of the recommendation system, and to perform information cocoon control on the recommendation system according to the information cocoon control strategy.

[0178] Specifically, the acquisition unit 501 in the device is used to acquire the number of positive feedback users, the number of negative feedback users, the total number of recommendations, and the intensity of free exploration of the recommended users in the recommendation system to be controlled by information cocoons, and to calculate the ratio between the number of negative feedback users and the number of positive feedback users, and to use the similarity calculation result between the number of positive feedback users and the total number of recommendations as the recommendation intensity based on the similarity between the users; the determination unit 502 is used to determine the state of the recommendation system based on the position of the ratio, the recommendation intensity based on the similarity between the users, and the intensity of free exploration of the users in the pre-acquired information cocoon phase transition diagram; the control unit 503 is used to determine the information cocoon control strategy corresponding to the state of the recommendation system, and to perform information cocoon control on the recommendation system according to the information cocoon control strategy.

[0179] The information cocoon control device provided in this invention acquires the number of positive feedback users, the number of negative feedback users, the total number of recommendations, and the intensity of free exploration by recommending users in a recommendation system to be controlled for information cocoons. It calculates the ratio between the number of negative feedback users and the number of positive feedback users, and uses the similarity calculation result between the number of positive feedback users and the total number of recommendations as the recommendation intensity based on user similarity. Based on the positions of the ratio, the recommendation intensity, and the intensity of free exploration by recommending users in a pre-acquired information cocoon phase transition diagram, it determines the state of the recommendation system. It then determines the information cocoon control strategy corresponding to the state of the recommendation system and performs information cocoon control on the recommendation system according to the strategy. This reveals the formation mechanism of information cocoons and fundamentally solves the information cocoon problem.

[0180] The embodiments of the information cocoon control device provided in this invention can be used to execute the processing flow of the above-described method embodiments. Its functions will not be repeated here, but can be referred to the detailed description of the above-described method embodiments.

[0181] Figure 6 This is a schematic diagram of the physical structure of an electronic device provided in an embodiment of the present invention, such as... Figure 6 As shown, the electronic device includes: a processor 601, a memory 602, and a bus 603;

[0182] The processor 601 and the memory 602 communicate with each other via the bus 603.

[0183] The processor 601 is used to call program instructions in the memory 602 to execute the methods provided in the above-described method embodiments, including, for example:

[0184] The system obtains the number of positive feedback users, the number of negative feedback users, the total number of recommendations, and the intensity of free exploration of the recommended users in the recommendation system to be subject to information cocoon control. It also calculates the ratio between the number of negative feedback users and the number of positive feedback users, and uses the similarity calculation result between the number of positive feedback users and the total number of recommendations users as the recommendation intensity based on the similarity of the recommended users.

[0185] The state of the recommendation system is determined based on the positions of the ratio, the recommendation strength based on the similarity of the recommended users, and the free exploration strength of the recommended users in the pre-acquired information cocoon phase transition diagram.

[0186] Determine the information cocoon control strategy corresponding to the state of the recommendation system, and perform information cocoon control on the recommendation system according to the information cocoon control strategy.

[0187] This embodiment discloses a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions, and when the program instructions are executed by a computer, the computer can perform the methods provided in the above-described method embodiments, such as:

[0188] The system obtains the number of positive feedback users, the number of negative feedback users, the total number of recommendations, and the intensity of free exploration of the recommended users in the recommendation system to be subject to information cocoon control. It also calculates the ratio between the number of negative feedback users and the number of positive feedback users, and uses the similarity calculation result between the number of positive feedback users and the total number of recommendations users as the recommendation intensity based on the similarity of the recommended users.

[0189] The state of the recommendation system is determined based on the positions of the ratio, the recommendation strength based on the similarity of the recommended users, and the free exploration strength of the recommended users in the pre-acquired information cocoon phase transition diagram.

[0190] Determine the information cocoon control strategy corresponding to the state of the recommendation system, and perform information cocoon control on the recommendation system according to the information cocoon control strategy.

[0191] This embodiment provides a computer-readable storage medium storing a computer program that causes the computer to execute the methods provided in the above-described method embodiments, including, for example:

[0192] The system obtains the number of positive feedback users, the number of negative feedback users, the total number of recommendations, and the intensity of free exploration of the recommended users in the recommendation system to be subject to information cocoon control. It also calculates the ratio between the number of negative feedback users and the number of positive feedback users, and uses the similarity calculation result between the number of positive feedback users and the total number of recommendations users as the recommendation intensity based on the similarity of the recommended users.

[0193] The state of the recommendation system is determined based on the positions of the ratio, the recommendation strength based on the similarity of the recommended users, and the free exploration strength of the recommended users in the pre-acquired information cocoon phase transition diagram.

[0194] Determine the information cocoon control strategy corresponding to the state of the recommendation system, and perform information cocoon control on the recommendation system according to the information cocoon control strategy.

[0195] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0196] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0197] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0198] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0199] In the description of this specification, the references to terms such as "an embodiment," "a specific embodiment," "some embodiments," "for example," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0200] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for controlling information cocoons, characterized in that, include: The system obtains the number of positive feedback users, the number of negative feedback users, the total number of recommendations, and the intensity of free exploration of the recommended users in the recommendation system to be subject to information cocoon control. It also calculates the ratio between the number of negative feedback users and the number of positive feedback users, and uses the similarity calculation result between the number of positive feedback users and the total number of recommendations users as the recommendation intensity based on the similarity of the recommended users. The state of the recommendation system is determined based on the positions of the ratio, the recommendation strength based on the similarity of the recommended users, and the free exploration strength of the recommended users in the pre-acquired information cocoon phase transition diagram. Determine the information cocoon control strategy corresponding to the state of the recommendation system, and perform information cocoon control on the recommendation system according to the information cocoon control strategy.

2. The information cocoon control method according to claim 1, characterized in that, The step of determining the state of the recommendation system based on the positions of the ratio, the recommendation strength based on the similarity of the recommended users, and the free exploration strength of the recommended users in the pre-acquired information cocoon phase transition diagram includes: If the location is determined to be within a first spatial region corresponding to diversification, then the recommendation system state is determined to be a diversification state. If the location is determined to be in the second spatial region corresponding to a mild information cocoon, then the recommendation system state is determined to be a mild information cocoon state. If the location is determined to be in a third spatial region corresponding to the deep information cocoon, then the recommendation system state is determined to be the deep information cocoon state.

3. The information cocoon control method according to claim 2, characterized in that, The determination of the information cocoon control strategy corresponding to the state of the recommendation system includes: If the recommendation system is determined to be in the state of deep information cocoon, and the ratio is lower than a preset ratio threshold, and the intensity of free exploration by recommended users is lower than a preset free exploration intensity threshold, then the information cocoon control strategy is determined to be to increase the utilization rate of negative samples and enhance the intensity of free exploration by users. If the recommendation system is determined to be in the state of the deep information cocoon, and the recommendation strength based on user similarity is higher than a preset similarity recommendation strength threshold, while the recommendation strength based on user free exploration is lower than a preset free exploration strength threshold, then the information cocoon control strategy is determined to be to suppress the use of similarity recommendations and enhance the intensity of user free exploration.

4. The information cocoon control method according to any one of claims 1 to 3, characterized in that, Obtaining the information cocoon phase transition diagram includes: Obtaining the information cocoon phase transition diagram includes: Data measurement results are obtained based on user interaction data; the data measurement results include the correlation strength between the recommendation distribution information entropy and the positive feedback ratio of observed users, the negative feedback ratio of observed users, the recommendation strength of observed users similarity, and the intensity of observed users' free exploration, as well as the global popularity of different item types and the correlation matrix between different item types; Based on the observation preference vector and correlation matrix of the observed users determined by the global popularity, the similarity between the observation preference vector and the item feature vector is calculated. Based on the similarity between the observation preference vector and the item feature vector and the recommendation strength of the observation user similarity, the recommendation probability of recommending each item to the observed user is determined, and each item is recommended to the observed user with the recommendation probability to obtain a dataset. The similarity between the observed user's inherent preference vector and the item feature vector, determined based on the global popularity, is used as the probability of accepting the recommendation. Positive feedback items with a probability greater than the probability of accepting the recommendation are selected from the dataset as positive feedback samples, and the remaining items are used as negative feedback samples. A stochastic differential equation is constructed based on the proportion of positive feedback from observed users, the proportion of negative feedback from observed users, positive feedback samples and negative feedback samples, the interaction function between the observed preference vector and the item feature vector, the intensity of free exploration by observed users, and the standard Wiener process, and the observed preference vector is updated based on the stochastic differential equation. The Falk-Planck equation is obtained from the stochastic differential equation, and the derivative of the Falk-Planck equation with respect to time is obtained to obtain the recommendation information entropy distribution. The relative information entropy distribution is calculated based on the recommended information entropy distribution and the inherent information entropy distribution. Based on the relative information entropy distribution, a first spatial region corresponding to diversification, a second spatial region corresponding to light information cocoons, and a third spatial region corresponding to deep information cocoons are determined. Relevant parameters of the information cocoon phase transition diagram are obtained through simulation. The information cocoon phase transition diagram is drawn based on the relevant parameters, the first spatial region, the second spatial region, and the third spatial region.

5. The information cocoon control method according to claim 4, characterized in that, The step of obtaining data measurement results based on user interaction data includes: A dataset is constructed based on recommendation logs and feedback logs from active users; the dataset includes item types corresponding to the items. From the dataset, we obtain behavioral features that indicate liking for recommended items to obtain positive feedback behavior; from the dataset, we obtain behavioral features that indicate disliking for recommended items to obtain negative feedback behavior. The ratio of positive feedback behavior to recommendation behavior is used as the positive feedback ratio of observed users, and the ratio of negative feedback behavior to recommendation behavior is used as the negative feedback ratio of observed users; the number of recommendation behavior is the sum of the number of positive feedback behavior and the number of negative feedback behavior. Establish the first correlation strength between the recommendation information entropy distribution and the proportion of positive feedback from observed users, and establish the second correlation strength between the recommendation information entropy distribution and the proportion of negative feedback from observed users.

6. The information cocoon control method according to claim 5, characterized in that, The step of obtaining data measurement results based on user interaction data includes: The average value of the positive feedback behavior distribution is used as the global popularity of different item types; Establish a third correlation strength between the recommendation information entropy distribution and global popularity.

7. The information cocoon control method according to claim 5, characterized in that, The step of obtaining data measurement results based on user interaction data includes: The covariance of the positive feedback behavior distribution is used as the correlation matrix between different item types; Establish a fourth correlation strength between the entropy distribution of recommendation information and the correlation matrix.

8. An information cocoon control device, characterized in that, include: The acquisition unit is used to acquire the number of positive feedback users, the number of negative feedback users, the total number of recommendations, and the intensity of free exploration of the recommended users in the recommendation system to be controlled by information cocoon, and to calculate the ratio between the number of negative feedback users and the number of positive feedback users, and to use the similarity calculation result between the number of positive feedback users and the total number of recommendations as the recommendation intensity of the similarity of the recommended users. The determining unit is used to determine the state of the recommendation system based on the ratio, the recommendation strength of the similarity between the recommended users, and the position of the free exploration strength of the recommended users in the pre-acquired information cocoon phase transition diagram; The control unit is used to determine the information cocoon control strategy corresponding to the state of the recommendation system, and to perform information cocoon control on the recommendation system according to the information cocoon control strategy.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images