Intelligent Recommendation Method, Device and Medium Based on Hybrid Modal Opinion Consensus Decision Making

Abstract

The present invention discloses an intelligent recommendation method, device and medium based on hybrid modal opinion consensus decision-making. The steps of the method include: collecting numerical opinions and linguistic opinions of multiple users on a target problem to form a hybrid opinion set; searching the user trust sets of each user and calculating the trust weights between users; converting numerical opinions into linguistic opinions and converting linguistic opinions into numerical opinions; evolving the converted numerical opinions and linguistic opinions; through repeated iteration of hybrid opinion conversion, trust weight calculation and hybrid opinion evolution until consensus is reached, and determining the recommended result output according to the consensus state. By establishing a mutual conversion mechanism between user numerical opinions and linguistic opinions, the present invention realizes the mutual conversion between opinions, can process numerical and linguistic opinions simultaneously, establishes an efficient and flexible hybrid opinion evolution framework, and finally generates accurate personalized recommendation results through group consensus, realizing efficient and accurate intelligent recommendation.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent recommendation, and in particular to an intelligent recommendation method, device and medium based on hybrid-modal opinion consensus decision-making. Background Art

[0002] In the current digital economy era, users are faced with a vast amount of choices of goods and services. By using data analysis, machine learning algorithms, etc., to predict users' ratings of goods or services, so as to accurately recommend goods or services to users and achieve intelligent recommendation of goods or services, which can greatly improve the user experience. Intelligent recommendation needs to analyze a large amount of historical evaluation data of users, and then screen out the goods or services with the highest evaluations. However, the ways in which users express their opinions and ratings on goods or services may be different. For example, some users may express their opinions by numerical ratings (such as 0 to 10 points), while some users may use language labels (such as "good review", "medium review", "bad review", etc.). Different modal opinion expression formats are difficult to directly compare or fuse, resulting in difficulty in directly judging users' preferences accurately based on different modal evaluations of goods or services by users.

[0003] The evolution of group opinions and consensus decision-making is that in a social network or group interaction, individual opinions gradually form a group consensus through mutual influence and adjustment, and finally reach a unanimous opinion through group interaction and opinion adjustment. Based on the group opinion evolution model, the recommendation results can be optimized by analyzing the opinion differences and interaction relationships between users. Based on the consensus decision-making method, by simulating group interaction and opinion adjustment, more accurate recommendation results can be provided for the intelligent recommendation system. In the prior art, the group opinion evolution model usually adopts the bounded confidence model. However, the traditional bounded confidence model assumes that individuals will only interact when the difference in their opinions is less than a certain trust threshold, so it can only be applied to opinions in a single expression format and cannot handle heterogeneous hybrid-modal opinion expression formats (numerical and linguistic). The mixed expression formats of numerical and linguistic opinions cannot be effectively converted and fused within the trust mechanism framework, making it impossible to fully utilize the group consensus decision-making mechanism to accurately capture users' preferences and ensure the accuracy and reliability of recommendations. Summary of the Invention

[0004] The technical problem to be solved by the present invention lies in: aiming at the technical problems existing in the prior art, the present invention provides an intelligent recommendation method, device and medium based on hybrid-modal opinion consensus decision-making with a simple implementation method, low cost, high accuracy and reliability, which can fully fuse the numerical and linguistic hybrid-modal opinions of users to reach a consensus and achieve accurate intelligent recommendation.

[0005] To solve the above technical problems, the technical solution proposed by the present invention is:

[0006] An intelligent recommendation method based on hybrid-modal opinion consensus decision-making, the steps include:

[0007] Collect numerical opinions and linguistic opinions of multiple users on the target problem to form a hybrid opinion set; the numerical opinions are values representing the degree of users' emotional tendency towards the target problem using numbers, and the linguistic opinions are types representing the emotional tendency of users towards the target problem using texts.

[0008] Hybrid opinion transformation: Convert the numerical opinions in the hybrid opinion set into linguistic opinions and convert the linguistic opinions into numerical opinions, respectively obtaining the converted numerical opinion set and linguistic opinion set; when converting numerical opinions into linguistic opinions, convert the numerical opinions expressed by other users into linguistic opinions to form a unified linguistic opinion format; when converting linguistic opinions into numerical opinions, convert the linguistic opinions expressed by other users into numerical opinions to form a unified numerical opinion format.

[0009] Trust weight calculation: Search for the user trust set of each user according to the current hybrid opinion set, where the user trust set expressing numerical opinions contains numerical opinions whose absolute value of the difference in numerical opinions of the user at different times is within the numerical trust threshold, and the user trust set expressing linguistic opinions contains linguistic opinions whose absolute value of the difference in linguistic opinions of the user at different times is within the linguistic trust threshold, and calculate the trust weights between users according to the user trust sets of each user.

[0010] Hybrid opinion evolution: Evolve the converted numerical opinion set and linguistic opinion set. During the evolution process, use the numerical opinions and linguistic opinions of the previous moment and the trust weights between the users to calculate the numerical opinions and linguistic opinions of the next moment respectively. After the evolution is completed, a new hybrid opinion set is formed.

[0011] Iteratively execute the hybrid opinion transformation, trust weight calculation, and hybrid opinion evolution until the linguistic opinions and numerical opinions in the iteratively obtained hybrid opinion set reach a consensus, and determine the recommended result output according to the current consensus state.

[0012] Further, the expression for converting the numerical opinion into a linguistic binary tuple is:

[0013]

[0014] where represents the deviation between the converted linguistic opinion of the user and the numerical opinion, , represents a function regarding the numerical opinion , represents the numerical opinion The transformed language binary tuple represents the user's language opinion, represents the approximate rounding function, represents the user's language granularity, ;

[0015] The expression for converting the language binary tuple into a numerical opinion is as follows:

[0016]

[0017] where, represents the inverse function with respect to the numerical opinion.

[0018] Furthermore, the search for the user trust set of each user according to the current mixed opinion set includes:

[0019] The user corresponding to the numerical opinion According to his own t numerical opinion and numerical trust threshold at the moment finds t the user trust set at the moment :

[0020]

[0021] where, represents the set of numerical opinions of all users found for the user at the moment t ; represents the found user, represents the user at the moment t the numerical opinion of, represents the user at the moment t the numerical opinion of, where ; represents the numerical trust threshold, ;

[0022] The user corresponding to the language opinion According to his own t language opinion and language trust threshold at the moment finds t the user trust set at the moment :

[0023]

[0024] where, represents the set of language opinions of all users found for the user at the moment tThe set of language opinions at a moment, represents the user at t the language opinion at the moment, represents the user at t the language opinion at the moment, where , represents the language trust threshold, , represents the language granularity of the user.

[0025] Furthermore, in calculating the trust weights between users according to the user trust sets of each user, the trust weight of user for user is calculated according to the following formula :

[0026]

[0027] where, represents the number of elements in the trust set of other users for their own opinions when user expresses a mixed opinion at t the moment, represents the self-trust value of user , represents the mixed opinion set of the user at t the moment, represents the trust set the total number of users within.

[0028] Furthermore, evolving the converted numerical opinion set and language opinion set includes:

[0029] Using the numerical opinion evolution model under limited confidence to evolve the converted numerical opinions, and the numerical opinion evolution model is:

[0030]

[0031] where, , respectively represent the numerical opinions of numerical opinion user at , t the moment, respectively represent the numerical opinions of numerical opinion user at t the moment, represents the trust weight of numerical opinion user for user , represents numerical opinion user The self - trust value, represents the total number of users in the trust set of other users for their own opinions when a numerical opinion is expressed by a numerical - opinion user at t the moment;

[0032] The converted numerical opinions are evolved using a language opinion evolution model under bounded confidence, and the language opinion evolution model is:

[0033]

[0034] where , represents the language opinion of a language - opinion user at , t the moment, represents the language opinion of user at t the moment, represents the trust weight of a language - opinion user for user , represents the self - trust value of a language - opinion user , represents the total number of users in the trust set of other users for their own opinions when a language opinion is expressed by a language - opinion user at t the moment.

[0035] Furthermore, the iteration of performing the hybrid opinion conversion, trust weight calculation, and hybrid opinion evolution until the language opinions and numerical opinions in the obtained hybrid opinion set reach a consensus includes:

[0036] Obtaining the hybrid opinion set obtained after each iteration;

[0037] If the converted numerical opinion set obtained satisfies the opinion set , and and the converted language opinion set satisfies the opinion set , and , where ~ represent each numerical opinion, ~ represent each language opinion, represents the number of numerical opinions or language opinions, and if the deviation of the iteration result value is within a preset allowable range, it is determined that a consensus has been reached.

[0038] An intelligent recommendation device based on hybrid - modality opinion consensus decision - making includes:

[0039] A mixed opinion collection module for collecting numerical opinions and linguistic opinions of multiple users on a target issue to form a mixed opinion set; the numerical opinion uses a numerical value to represent the degree of the user's emotional tendency towards the target issue, and the linguistic opinion uses text to represent the type of the user's emotional tendency towards the target issue;

[0040] A mixed opinion conversion module for converting numerical opinions in the mixed opinion set into linguistic opinions and converting linguistic opinions into numerical opinions to obtain the converted numerical opinions and linguistic opinions;

[0041] A trust weight calculation module for searching the user trust set of each user according to the current mixed opinion set, where the user trust set expressing numerical opinions contains numerical opinions whose absolute value of the difference in numerical opinions of the user at different times is within the numerical trust threshold, and the user trust set expressing linguistic opinions contains linguistic opinions whose absolute value of the difference in linguistic opinions of the user at different times is within the linguistic trust threshold, and calculating the trust weight between users according to the user trust sets of each user;

[0042] A mixed opinion evolution module for evolving the converted numerical opinion set and linguistic opinion set. During the evolution process, the numerical opinions and linguistic opinions of the previous moment and the trust weights between the users are used to calculate the numerical opinions and linguistic opinions of the next moment correspondingly. After the evolution is completed, a new mixed opinion set is formed;

[0043] An iterative output module for iteratively executing the mixed opinion conversion, trust weight calculation, and mixed opinion evolution until the linguistic opinions and numerical opinions in the iteratively obtained mixed opinion set reach a consensus, and determining the recommended result output according to the current consensus state.

[0044] An electronic device includes a processor and a memory. The memory is used to store a computer program, and the processor is used to execute the computer program to execute the method as described above.

[0045] A computer-readable storage medium storing a computer program, where the computer program, when executed by a processor, implements the method as described above.

[0046] Compared with the prior art, the advantages of the present invention are as follows: By establishing a mutual conversion mechanism between user numerical opinions and language opinions, the present invention realizes the mutual conversion between numerical opinions and language opinions, calculates the trust weights between users using the user trust set, evolves the mixed opinion set respectively based on the calculated trust weights to form a new mixed opinion set, and through repeated iteration to update the trust set and the opinion evolution process until the group opinions reach a consensus, thereby determining the intelligent recommendation result. It can process numerical opinions and language opinions simultaneously, establish an efficient and flexible mixed opinion evolution framework, effectively address the divergence problem between numerical opinions and language opinions, enable individuals with different expression formats to communicate effectively, and at the same time, based on the consensus mechanism, help the group form a consistent decision under multimodal opinion expression, so as to finally generate accurate personalized recommendation results through group consensus, achieve efficient and accurate intelligent recommendation, and improve the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic diagram of the implementation process of the intelligent recommendation method based on the mixed-modal opinion consensus decision in this embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] The present invention will be further described below in conjunction with the accompanying drawings of the specification and specific preferred embodiments, but the protection scope of the present invention is not limited thereby.

[0049] Taking the intelligent e-commerce recommendation system as an example, the intelligent e-commerce recommendation system can help users discover interesting products or services. In the prior art, such recommendation systems usually only rely on the numerical ratings of users (such as star ratings, purchase volumes, browsing durations, etc.) or only rely on the language tags in user reviews (such as "good review", "medium review", "bad review", etc.). However, in reality, numerical ratings and language tags often coexist. A single data expression method cannot fully reflect the complex preferences of users. Especially in a diverse user group, the individual opinion expression methods may vary greatly, and the recommendation results determined directly based on a single type of evaluation data are not accurate. In addition, the discrete language opinions expressed by individuals are prone to result in fuzzy and inaccurate results, while the continuous numerical opinions also have uncertainties. Both single numerical opinions and language opinions lack certainty, and there are differences between discrete language preferences and continuous numerical preferences, making it difficult to directly establish a corresponding opinion conversion and mixed opinion expression fusion evolution mechanism between the two.

[0050] The present invention first collects numerical opinions and linguistic opinions of users on problems related to a certain product or service to form an initial mixed opinion set, and at the same time establishes a mutual conversion mechanism between numerical opinions and linguistic opinions to achieve the mutual conversion between numerical opinions and linguistic opinions. Then, by finding out the user trust set to calculate the trust weights between users, based on the calculated trust weights, the numerical opinions and linguistic opinions in the mixed opinion set are evolved respectively to form a new mixed opinion set. After repeated iteration and update of the trust set and the opinion evolution process until the group opinions reach a consensus, and then determine the recommendation result for the product or service. It can process numerical opinions and linguistic opinions at the same time, establish an efficient and flexible mixed opinion evolution framework, effectively address the divergence problem between numerical opinions and linguistic opinions, enable individuals with different expression formats to communicate effectively, and at the same time, based on the consensus mechanism, help the group form a consistent decision under the multi-modal opinion expression, so as to finally generate accurate personalized recommendation results through group consensus, achieve efficient and accurate intelligent recommendation, and improve the user experience.

[0051] Taking the intelligent e-commerce recommendation system as an example, there are a large number of opinions of users on products in different ways in the evaluation details of the e-commerce platform: one is numerical scoring, for example, rating a certain electronic product 4 stars; the other is providing linguistic comments, such as: "I like this mobile phone very much. Its performance is very good. I recommend it for purchase!" and so on. By adopting the method of the present invention, the numerical scores and linguistic comments of all users can be mutually converted and integrated together to form a comprehensive group opinion. If the linguistic opinion and the numerical opinion finally reach a consensus, the consensus decision result will be used as the corresponding recommendation result. For example, if the group rating of a certain product is 4.5 stars and the consensus in the linguistic comments is "like very much", then this product will be recommended with high priority; correspondingly, if the group opinion of a certain product is negative after reaching a consensus, the recommendation for this product will be reduced accordingly.

[0052] As Figure 1 shown, the detailed steps of the intelligent recommendation method based on the mixed-modal opinion consensus decision in this embodiment include:

[0053] Step S01. Collect numerical opinions and linguistic opinions of multiple users on the target problem to form a mixed opinion set; the numerical opinion is a value representing the degree of emotional tendency of the user on the target problem using a numerical value, and the linguistic opinion is a type representing the emotional tendency of the user on the target problem using text.

[0054] In this embodiment, the numerical opinion and the linguistic opinion range are first defined, the opinions of the same batch of users on problems such as a certain product or service are collected to form an initial mixed opinion set, and a self-confidence value is set for the users.

[0055] Specifically, the numerical opinion score can be defined as 0-10 points (configurable), and the linguistic opinion range ("good review", "medium review", "bad review") is , the language granularity is . Collect the numerical opinion scores and language opinions of the same target user group , with the group size being . Uniformly store them in the database, ensuring the uniformity and integrity of the data format, and indicating the type (numerical or linguistic) of each user's feedback opinion. At the same time, clean them to remove invalid or incomplete feedback, and organize to form the user's initial mixed opinion set , where represents the numerical opinion of user at time , and represents the language opinion of user at time . Set the self-confidence value of user

[0056] . Specifically, the self-confidence values of all users can be set to be the same as

[0057] . It can be understood that both the numerical opinion scores and the language opinion ranges can be configured according to actual needs. For example, the language opinion range can also be configured as "agree", "neutral", "disagree", etc.

[0058] In this embodiment, converting the language opinion to a numerical opinion means that users expressing numerical opinions can only fuse numerical opinions. Therefore, the language opinions expressed by other users need to be converted into numerical opinions to form a unified numerical opinion format, so that the numerical opinions of each user can be fused for opinion update later; converting the numerical opinion to a language opinion means that users expressing language opinions can only fuse language opinions. Therefore, the numerical opinions expressed by other users need to be converted into language opinions to form a unified language opinion format, so that the language opinions of each user can be fused for opinion update later.

[0059] In this embodiment, by establishing a mutual conversion mechanism between the user's numerical opinions and language opinions, the mutual conversion between numerical and language opinions is realized. The numerical opinions are converted into language opinions, and at the same time, the language opinions are also converted into numerical opinions, so that the two can be compared and fused under a unified framework, so that individuals with different opinion expression formats can better communicate with each other. For example, language labels such as "agree" can be mapped to numerical intervals, and numerical scores can also be converted into language labels.

[0060] Specifically, based on the tuple model, the language granularity can be introduced to construct a first conversion model for converting numerical opinions into linguistic opinions. For example, the following first conversion model can be used to convert numerical opinion data into linguistic tuples , The initial value of which is 0, that is, the linguistic tuple :

[0061] (1)

[0062] wherein represents the deviation between the converted linguistic opinion of the user and the numerical opinion, , represents a function with respect to the numerical opinion , represents the numerical opinion converted into a linguistic tuple, represents the linguistic opinion of the user, represents the rounding function, represents the language granularity of the user, .

[0063] The tuple model is a mathematical model for dealing with uncertainty and ambiguity by representing fuzzy information or fuzzy language as a tuple. Each tuple consists of a linguistic term and a numerical value, and the numerical value ranges between (-0.5, 0.5). Through this model, continuous representation can be allowed in the domain of linguistic information, so that any information can be represented during the aggregation process. In this embodiment, by constructing the above first conversion model based on the tuple model, the high precision (floating decimals between 0 and 10) of numerical opinions is mapped into a discretized form of linguistic opinions, which can better quantify and integrate the conversion of individual numerical opinions into linguistic opinions.

[0064] As shown in Equation (1), the first conversion model, by using corresponding to the linguistic opinion , the numerical opinions can be mapped to discrete linguistic labels, enabling the fuzzy numerical information to match the predefined linguistic labels. By adopting the rounding method, the mapping process can be simplified, ignoring the high-precision details of unimportant numerical opinions, avoiding over-refinement of the numerical values, and paying more attention to the relative strength of these opinions. Thus, it is convenient to quickly map them to a coarser linguistic category. At the same time, through the rounding operation, the numerical opinions can be effectively mapped to a fixed interval of linguistic labels. For example, rounding the numerical value 2.5 to 3, it can be mapped to the "neutral" category, and the matching between the numerical value and the linguistic label after the rounding operation is more intuitive and clear. At the same time, by introducing the user's linguistic granularity Normalizing the numerical opinion values can effectively map the numerical opinions to the standardized values in the interval [0, 1].

[0065] Furthermore, in order to convert the linguistic labels in the linguistic pair back to the corresponding numerical interval and determine the specific numerical score to complete the reverse mapping, according to the inverse function of the conversion model that converts numerical opinions into linguistic opinions, a second conversion model that converts linguistic opinions into numerical opinions can be constructed to realize the conversion of linguistic opinions into numerical opinions, that is, the process of converting a linguistic pair into a numerical opinion is the inverse function of converting a numerical opinion into a linguistic pair. For example, based on the fuzzy set theory and the principle of membership (numerical) function, the following second conversion model can be used to convert the linguistic pair into numerical opinions :

[0066] (2)

[0067] Among them, represents the inverse function regarding the numerical opinion.

[0068] According to the above formula, the subscript of the linguistic opinion can be mapped to a specific numerical value for calculation, where is 0 when the linguistic opinion is first converted into a numerical opinion, and new will be generated after the first evolution, that is, calculate according to for calculation.

[0069] When dealing with the standardization between different language labels and the comparison between numerical values, the traditional binary tuple model may encounter problems of inconsistency or incomparability. In this embodiment, a first conversion model for converting numerical opinions into language opinions is first constructed based on the binary tuple model, and then a second conversion model for converting language opinions into numerical opinions is constructed based on the inverse function of the first conversion model. By using language labels and membership degrees to restore or infer numerical opinions, the language and numerical opinions can be standardized and normalized, unifying the scales of numerical opinions and language opinions, enabling the two to be compared and converted within the same framework, thus adapting to the complexity of multi-modal opinions.

[0070] By adopting the above mutual conversion mechanism between user numerical opinions and language opinions, this embodiment can break the limitations of only using single numerical opinions or language opinions traditionally, enabling different types of opinions to be converted into each other under appropriate mechanisms, thus efficiently handling the evolution of mixed opinions among groups.

[0071] Step S03. Trust weight calculation: Search for the user trust sets of each user according to the current mixed opinion set. Among them, the user trust set expressing numerical opinions contains numerical opinions whose absolute values of the differences in numerical opinions of the user at different times are within the numerical trust threshold range, and the user trust set expressing language opinions contains language opinions whose absolute values of the differences in language opinions of the user at different times are within the language trust threshold range. Calculate the trust weights between users according to the user trust sets of each user.

[0072] In this embodiment, by setting the language trust threshold, the user trust set is found, and then the trust weight of a user in other users is calculated.

[0073] Specifically, searching for the user trust sets of each user according to the current mixed opinion set includes:

[0074] The user corresponding to the numerical opinion According to his own t Numerical opinion at a certain moment and the numerical trust threshold Find out t The user trust set at a certain moment :

[0075] (3)

[0076] Among them, Represents the set of numerical opinions of all users found for user At a certain moment, t The set of numerical opinions, Represents the user found, Represents user At t The numerical opinion at a certain moment, Represents user The numerical opinion at t time, where ; represents the numerical trust threshold, .

[0077] The user corresponding to the linguistic opinion According to his own t linguistic opinion and linguistic trust threshold at finds out t the user trust set at :

[0078] (4)

[0079] Among them, represents the set of linguistic opinions of all users found for user at t time, represents the linguistic opinion of user at t time, represents the linguistic opinion of user at t time, where , represents the linguistic trust threshold, , represents the linguistic granularity of the user.

[0080] As an optional implementation manner, when calculating the trust weights between users according to the user trust sets of each user, a trust weight allocation method based on the HK model can be used, and the trust weight of user for user is calculated according to the following formula:

[0081] (5)

[0082] Among them, represents the number of elements in the trust set of other users' opinions on their own opinions when user expresses a mixed opinion at t time in represents the self-trust value of user , represents the mixed opinion set of the user at t time, represents the total number of users in the trust set in, and user is a certain user in the mixed opinion set. The above trust weight calculation method is applicable to the users corresponding to numerical opinions and the users corresponding to linguistic opinions.

[0083] When the same group of users express different opinions on the same event, each user has a certain self - confidence value for themselves and also has a certain trust value for other users except their own opinions. In this embodiment, a trust weight calculation model is constructed based on the trust weight allocation method of the HK model. Using this calculation model, the trust weights for the opinions of other users except the self - confidence value of the user for their own opinions can be evenly distributed, and the trust weight value of each user for other users can be effectively calculated.

[0084] Step S04. Hybrid opinion evolution: Evolve the converted numerical opinion set and the linguistic opinion set. During the evolution process, use the numerical opinions and linguistic opinions at the previous moment and the trust weights between users to calculate the numerical opinions and linguistic opinions at the next moment. After the evolution is completed, a new hybrid opinion set is formed.

[0085] In this embodiment, by separately constructing a numerical opinion evolution model under bounded confidence and a linguistic opinion evolution model under bounded confidence, the opinions of numerical opinion expressors are evolved according to the numerical opinion evolution model, and the opinions of linguistic opinion expressors are evolved according to the linguistic opinion evolution model, thereby forming a new hybrid opinion set.

[0086] Specifically, evolving the converted numerical opinions and linguistic opinions includes:

[0087] Evolve the converted numerical opinions using the numerical opinion evolution model under bounded confidence. The numerical opinion evolution model is:

[0088] (6)

[0089] Wherein, 、 respectively represent the numerical opinions of numerical opinion user at 、 t moments, respectively represent the numerical opinions of numerical opinion user at t moment, represents the trust weight of numerical opinion user for user , represents the self - confidence value of numerical opinion user , represents the total number of users in the trust set of other users for their own opinions when numerical opinion user expresses a numerical opinion at t moment;

[0090] The converted numerical opinions are evolved using the language opinion evolution model under limited confidence, and the language opinion evolution model is as follows:

[0091] (7)

[0092] wherein 、 represent the language opinions of the language opinion user at 、 t moments, represents the language opinion of user at t moment, represents the trust weight of the language opinion user in user , represents the self - trust value of the language opinion user , represents the total number of users in the trust set of other users for their own opinions when the language opinion user expresses the language opinion at t moment.

[0093] In this embodiment, by constructing the above - mentioned numerical opinion evolution model under limited confidence (Equation (6)), evolving after converting the language opinion into a numerical opinion, and constructing the language opinion evolution model under limited confidence (Equation (7)), evolving after converting the numerical opinion into a language opinion, and combining the numerical opinion and the language opinion at moment after each evolution is completed, a new mixed opinion set is formed and stored in the new mixed opinion set, which can evolve different opinion formats of users, can simulate the opinion evolution process under different trust levels, and enables the model to adapt to a complex social network environment.

[0094] Step S05. Iteratively execute step S02 for mixed opinion conversion, step S03 for trust weight calculation, and step S04 for mixed opinion evolution until the language opinions and numerical opinions in the iteratively obtained mixed opinion set reach a consensus, and determine the recommended result output according to the current consensus state.

[0095] In this embodiment, by continuously iteratively identifying the user trust set corresponding to the new mixed opinion set, calculating the trust weights between users, performing mixed opinion conversion according to step S02, and performing mixed opinion evolution according to step S04, if the mixed opinions reach a consensus, the recommended result output is determined according to the current consensus state.

[0096] Specifically, iteratively executing mixed opinion conversion, trust weight calculation, and mixed opinion evolution until the language opinions and numerical opinions in the iteratively obtained mixed opinion set reach a consensus includes:

[0097] Obtain the mixed opinion set obtained after each iteration execution;

[0098] If the mixed opinion set obtained is converted into a numerical opinion set and satisfies the opinion set , and and is converted into a linguistic opinion set and satisfies the opinion set , and , where ~ represents each numerical opinion, ~ represents each linguistic opinion, represents the number of numerical opinions or linguistic opinions, and if the deviation of the iteration result value is within the preset allowable range, it is determined that consensus is reached.

[0099] In this embodiment, by continuously iterating using the group opinion consensus mechanism, as more users participate in scoring and commenting, the system can update the group opinion in real time, further optimizing the recommendation result. If the original evaluation of a certain product or service is relatively divided, the recommendation strategy will be quickly adjusted through the above steps, enabling real-time accurate and reliable intelligent recommendation.

[0100] Through the above method, this embodiment can effectively analyze and convert the differences between numerical opinions and linguistic opinions, help the group to guide the realization of the unity of group opinions in the face of opinion differences, effectively solve the problem that traditional numerical scoring and linguistic comments cannot be directly integrated, and enable the effective comprehensive generation of more accurate and reliable recommendation results from different types of user feedback.

[0101] The present invention can be applied to the intelligent recommendation of products or services using e-commerce evaluation data, and can also be applied to the field of social media analysis and other fields with mixed opinion formats to achieve group opinion consensus decision-making. By adopting the above method of the present invention to mutually transform and evolve the numerical opinions and linguistic opinions of individuals, it can promote the interaction of individuals with different expression formats, and thus promote the formation of group opinion fusion consensus.

[0102] The following takes the implementation of the present invention to achieve the intelligent recommendation of a certain product in a specific application embodiment as an example to further illustrate the present invention. The detailed steps are as follows:

[0103] Step S1: Define the ranges of numerical opinions and linguistic opinions, collect the evaluation opinions of the same batch of users on a certain product to form an initial mixed opinion set, and set a confidence value for the users.

[0104] Specifically, define the numerical opinion score (0 - 10 points), and the linguistic opinion range ("good review", "medium review", "bad review"), that is , then the linguistic granularity is . Obtain the same target user group The numerical opinion scores and language opinions are uniformly stored in the database, and the group size is , ensuring that the data is in a unified format, complete, and indicating the feedback type (numerical or linguistic) of each user. After data cleaning to remove invalid or incomplete feedback, an initial mixed opinion set of users is formed . Among them, 0.4984, 0.5853, and 0.2238 are the numerical opinions of users at the moment, and the rest are the language opinions of users at the moment. At this time, the self-confidence values of all users are set , then the self-confidence value matrix is: .

[0105] Step S2: Convert the numerical opinions in the mixed opinion set into language opinions and convert the language opinions into numerical opinions to obtain the converted numerical opinions and language opinions.

[0106] Specifically, convert the numerical opinion in accordance with into a language binary tuple , and convert the language binary tuple in accordance with into a numerical opinion . According to the numerical opinions and language opinions of the mixed opinion expresser , convert the opinions respectively:

[0107] (1) For the language opinion expresser, convert the numerical opinion into a language opinion, that is:

[0108] Convert the numerical opinion corresponding to 0.4984 into a language opinion:

[0109] Convert the numerical opinion corresponding to 0.5853 into a language opinion:

[0110] Convert the numerical opinion corresponding to 0.2238 into a language opinion:

[0111] At this time, , ;

[0112] , ;

[0113] , .

[0114] Also , the deviation generated after converting the numerical opinion into a linguistic opinion is acceptable, which further verifies the rationality of the converted linguistic opinion.

[0115] Therefore, at the converted linguistic opinion at the moment is .

[0116] (2) For the numerical opinion expresser, convert the linguistic opinion into a numerical opinion, that is:

[0117] Convert the linguistic opinion corresponding to 2 into a numerical opinion: , The initial value is 0;

[0118] Convert the linguistic opinion corresponding to 1 into a numerical opinion: ;

[0119] Therefore, at the converted numerical opinion at the moment is .

[0120] Step S3: Set the numerical and linguistic trust thresholds respectively, find out the user trust set, and calculate the trust weights between users.

[0121] Set the numerical trust threshold , and calculate the linguistic trust threshold , then the calculated result is .

[0122] Furthermore, find out the user trust set. The numerical opinion expresser finds out the trust set at time t according to his numerical opinion at time t and the numerical trust threshold , and the linguistic opinion expresser finds out the trust set at time t according to his linguistic opinion at time t and the linguistic trust threshold .

[0123] S33: Calculate the trust weight (numerical opinion expresser or linguistic opinion expresser) of user for other users .

[0124] (1) For the linguistic opinion expresser, the linguistic opinion boundary trust value is , then:

[0125] Find out the user trust set of the linguistic opinion corresponding to 2: , and the trust weights assigned to other users are: ;

[0126] Find out the user trust set of the linguistic opinion corresponding to 1: , the trust weight distribution for other users is as follows: .

[0127] (2) For numerical opinion expressers, the trust value of the numerical opinion boundary is , then:

[0128] Find the user trust set corresponding to 0.4984: , the trust weight distribution for other individuals is as follows: ;

[0129] Find the user trust set corresponding to 0.5853: , the trust weight distribution for other individuals is as follows: ;

[0130] Find the user trust set corresponding to 0.2238: , the trust weight distribution for other individuals is as follows: .

[0131] S4. Respectively use the numerical opinion evolution model under limited confidence and the linguistic opinion evolution model under limited confidence to evolve the opinions of numerical opinion expressers according to the numerical opinion evolution model, and evolve the opinions of linguistic opinion expressers according to the linguistic opinion evolution model to form a new mixed opinion set.

[0132] First, convert the linguistic opinion into a numerical opinion and then evolve it: , convert the numerical opinion into a linguistic opinion and then evolve it: , and evolve respectively according to the numerical opinion evolution model and the linguistic opinion evolution model at to obtain the opinion values at the moment:

[0133] Opinions corresponding to 0.4984: ;

[0134] Opinions corresponding to 2: ;

[0135] Opinions corresponding to 1: ;

[0136] Opinions corresponding to 0.5853: ;

[0137] Opinions corresponding to 0.2238: ;

[0138] At moment, the new mixed opinion set formed is .

[0139] S5. Iteratively execute the conversion of mixed opinions, the calculation of trust weights, and the evolution of mixed opinions until the language opinions and numerical opinions in the obtained mixed opinion set reach a consensus, and determine the recommended result output according to the current consensus state.

[0140] According to the above steps of language opinion evolution, the language opinions go through the following 5 rounds of iteration to obtain the language opinion evolution results as shown in Table 1. The group language opinions are unified and reach a consensus at time.

[0141] Table 1: Process of Language Opinion Evolution

[0142]

[0143] Next, according to the above steps of numerical opinion evolution, the numerical opinions go through the following 5 rounds of iteration to obtain the numerical opinion evolution results as shown in Table 2. The opinion values of the group numerical opinions are approximately equal after rounding at time (rounding is allowed here for numerical opinions because the numerical opinions evolve iteratively over time and the final opinion values will be approximately equal).

[0144] Table 2: Process of Numerical Opinion Evolution

[0145]

[0146] Finally, the mixed opinions dynamically lead to five rounds of iteration, as shown in Table 3.

[0147] Table 3: Iteration Process of Mixed Opinions

[0148]

[0149] From Table 3, it can be obtained that the language opinions of the users reach a consensus at time, and the numerical opinions are approximately 0.65 and also reach a consensus at time, indicating that the users can gradually form consensus opinions during the continuous iteration process. Furthermore, the recommended results are determined according to the consensus decision results. Also, because the numerical opinions can be converted into the language opinion S1 at time, the opinions of all users in the mixed opinions reach a consensus as S1. S1 corresponds to a medium review, and the numerical value also tends to a medium state. Therefore, it is determined that the group evaluation of the product is a medium review, and a medium recommendation level is configured during the recommendation. As can be seen from the above, the present invention can fully integrate numerical opinions and language opinions for communication and integration, and use group opinion consensus decision to finally achieve accurate and reliable intelligent recommendation.

[0150] An intelligent recommendation device based on mixed-modal opinion consensus decision in this embodiment includes:

[0151] A mixed opinion collection module for collecting numerical opinions and language opinions of multiple users on a target problem to form a mixed opinion set; the numerical opinion is a value representing the degree of emotional tendency of a user on the target problem using a numerical value, and the language opinion is a type representing the emotional tendency of a user on the target problem using text;

[0152] A mixed opinion conversion module for converting the numerical opinions in the mixed opinion set into language opinions and converting the language opinions into numerical opinions to obtain the converted numerical opinions and language opinions;

[0153] A trust weight calculation module for searching the user trust set of each user according to the current mixed opinion set, the numerical opinions whose absolute value of the difference in numerical opinions of a user at different times is within the numerical trust threshold range, and the language opinions whose absolute value of the difference in language opinions of a user at different times is within the language trust threshold range, and calculating the trust weight between users according to the user trust set of each user;

[0154] A mixed opinion evolution module for evolving the converted numerical opinions and language opinions, and calculating the numerical opinions and language opinions of the next moment using the numerical opinions and language opinions of the previous moment and the trust weight between users during the evolution process, and forming a new mixed opinion set after the evolution is completed;

[0155] An iterative output module for iteratively executing the mixed opinion conversion, trust weight calculation, and mixed opinion evolution until the language opinions and numerical opinions in the iteratively obtained mixed opinion set reach a consensus, and determining the recommended result output according to the current consensus state.

[0156] The intelligent recommendation device based on mixed-modal opinion consensus decision in this embodiment corresponds one-to-one with the above intelligent recommendation method based on mixed-modal opinion consensus decision, and will not be elaborated here one by one.

[0157] This embodiment further provides an electronic device, including a processor and a memory, the memory is used for storing a computer program, and the processor is used for executing the computer program to execute the method as described above.

[0158] It can be understood that the above method of this embodiment can be executed by a single device, such as a computer or a server, etc., or can also be applied to a distributed scenario where multiple devices cooperate with each other to complete. In the case of a distributed scenario, one of the multiple devices can only execute one or more steps of the above method of this embodiment, and the multiple devices interact with each other to complete the above method. The processor can be implemented in ways such as a general-purpose CPU, a microprocessor, an application-specific integrated circuit, or one or more integrated circuits, etc., and is used to execute relevant programs to implement the above method of this embodiment. The memory can be implemented in forms such as a read-only memory ROM, a random access memory RAM, a static storage device, and a dynamic storage device, etc. The memory can store an operating system and other application programs. When implementing the above method of this embodiment through software or firmware, the relevant program codes are stored in the memory and are called and executed by the processor.

[0159] This embodiment further provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, it implements the method as described above.

[0160] Those skilled in the art should understand that the above embodiments of the present invention can provide a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes. The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the function specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks. These computer program instructions can also be stored in a computer-readable memory capable of guiding a computer or other programmable data processing devices to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured product including an instruction device, and the instruction device implements the function in Figure 1 one flow or multiple flows and / or blocks Figure 1The functions specified in one or more boxes. These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes and / or boxes Figure 1 One process or more processes and / or boxes Figure 1 Steps for implementing the functions specified in one or more boxes

[0161] The above are only the preferred embodiments of the present invention and do not impose any formal restrictions on the present invention. Although the present invention has been disclosed above with the preferred embodiments, it is not intended to limit the present invention. Therefore, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention shall fall within the scope of the technical solution of the present invention

Claims

1. An intelligent recommendation method based on hybrid modal opinion consensus decision-making, characterized in that the steps Including: Collecting numerical opinions and linguistic opinions of multiple users on a target problem to form a mixed opinion set; The numerical opinions are values using numbers to represent the degree of users' emotional tendency towards the target problem, and the linguistic opinions are types using texts to represent the emotional tendency of users towards the target problem; Mixed opinion conversion: Converting the numerical opinions in the mixed opinion set into linguistic opinions and converting the linguistic opinions into numerical opinions to obtain a converted numerical opinion set and a linguistic opinion set respectively; when converting numerical opinions into linguistic opinions, converting the numerical opinions expressed by other users into linguistic opinions to form a unified linguistic opinion format; when converting linguistic opinions into numerical opinions, converting the linguistic opinions expressed by other users into numerical opinions to form a unified numerical opinion format; Trust weight calculation: Searching for the user trust sets of each user according to the current mixed opinion set, where the user trust set of a user expressing numerical opinions contains numerical opinions whose absolute value of the difference in numerical opinions of different users at the same moment is within the numerical trust threshold range, and the user trust set of a user expressing linguistic opinions contains linguistic opinions whose absolute value of the difference in linguistic opinions of different users at the same moment is within the linguistic trust threshold range, and calculating the trust weights between users according to the user trust sets of each user; Mixed opinion evolution: Evolving the converted numerical opinion set and linguistic opinion set, and calculating the numerical opinions and linguistic opinions of the next moment using the numerical opinions and linguistic opinions of the previous moment and the trust weights between users during the evolution process, and forming a new mixed opinion set after the evolution is completed; Iteratively executing the mixed opinion conversion, trust weight calculation, and mixed opinion evolution until the linguistic opinions and numerical opinions in the iteratively obtained mixed opinion set reach a consensus, and determining the recommended result output according to the current consensus state.

2. The intelligent recommendation method based on hybrid modal opinion consensus decision-making according to claim 1, wherein Convert numerical opinions into linguistic pairs The expression is as follows: Among them, represents the deviation between the user's converted language opinion and numerical opinion, , represents a function regarding the numerical opinion . represents the numerical opinion converted into a language binary tuple, represents the user's language opinion, represents a rounding function, represents the user's language granularity, ; Convert the language pair into a numerical opinion The expression is: Among them, represents the inverse function of the numerical opinion.

3. The intelligent recommendation method based on hybrid modal opinion consensus decision-making according to claim 1, characterized in that The searching for the user trust sets of each user according to the current mixed opinion set includes: User corresponding to the numerical opinion According to oneself t The numerical opinion and numerical trust threshold at a certain moment Find out t The user trust set at a certain moment : Among them, represents the set of numerical opinions of all users found by the user at t time, represents the user found, represents the user at t time of numerical opinion, represents the user at t time of numerical opinion, where ; represents the numerical trust threshold, ; User corresponding to the language opinion Based on their own t Language opinion and language trust threshold at a given moment Find out t User trust set at a given moment : Among them, represents the set of language opinions of all users found by the user at t moment, represents the user at t moment of language opinion, represents the user at t moment of language opinion, where , represents the language trust threshold, , represents the user's language granularity.

4. The intelligent recommendation method based on hybrid modal opinion consensus decision-making according to claim 1, characterized in that In calculating the trust weights between users based on the user trust sets of each user, the trust weight of user for user is calculated according to the following formula : Among them, represents the number of elements in the trust set of other users' opinions on the user's own opinion when expressing mixed opinions at at t time, and represents the self-trust value of the user . represents the mixed opinion set of the user at t time, and represents the total number of users in the trust set 5. The intelligent recommendation method based on hybrid-modal opinion consensus decision-making according to claim 1, wherein The evolving the converted numerical opinion set and linguistic opinion set includes: Evolving the converted numerical opinions using a numerical opinion evolution model under bounded confidence, and the numerical opinion evolution model is: Among them, and respectively represent the numerical opinions of the user at and t moments. and respectively represent the numerical opinions of the user t at represents the trust weight of the numerical opinion user towards the user ; represents the self - trust value of the numerical opinion user ; represents the total number of users within the trust set of other users for the numerical opinion expressed by the numerical opinion user at t moment; Evolving the converted numerical opinions using a linguistic opinion evolution model under bounded confidence, and the linguistic opinion evolution model is: Among them, and represent the language opinions of the user at and t moments. represents the user at t moment's language opinion. represents the trust weight of the language opinion user towards the user represents the self-trust value of the language opinion user represents the total number of users in the trust set of other users' opinions on their own opinions when the language opinion user expresses the language opinion at t moment.​​ 6. The intelligent recommendation method based on hybrid modality opinion consensus decision according to any one of claims 1 to 5, characterized in that The iteratively executing the mixed opinion conversion, trust weight calculation, and mixed opinion evolution until the linguistic opinions and numerical opinions in the iteratively obtained mixed opinion set reach a consensus includes: Obtaining the mixed opinion set obtained after each iterative execution; If the obtained mixed opinion set is converted into a numerical opinion set and meets the opinion set , and and when it is converted into a linguistic opinion set and meets the opinion set , and , where ~ represent respective numerical opinions, ~ represent respective linguistic opinions, represents the quantity of numerical opinions or linguistic opinions, and if the deviation of the iterative result value is within the preset allowable range, it is determined that consensus is reached.

7. An intelligent recommendation device based on hybrid-modal opinion consensus decision-making, characterized in that, Including: A mixed opinion collection module for collecting numerical opinions and linguistic opinions of multiple users on a target problem to form a mixed opinion set; The numerical opinions are values using numbers to represent the degree of users' emotional tendency towards the target problem, and the linguistic opinions are types using texts to represent the emotional tendency of users towards the target problem; A mixed opinion conversion module is used to convert the numerical opinions in the mixed opinion set into linguistic opinions and convert the linguistic opinions into numerical opinions, respectively obtaining a converted numerical opinion set and a linguistic opinion set; when converting numerical opinions into linguistic opinions, it converts the numerical opinions expressed by other users into linguistic opinions to form a unified linguistic opinion format; when converting linguistic opinions into numerical opinions, it converts the linguistic opinions expressed by other users into numerical opinions to form a unified numerical opinion format. A trust weight calculation module is used to search for the user trust sets of each user according to the current mixed opinion set. The user trust set expressing numerical opinions contains numerical opinions within the numerical trust threshold range of the absolute value of the numerical opinion differences of different users at the same moment, and the user trust set expressing linguistic opinions contains linguistic opinions within the linguistic trust threshold range of the absolute value of the linguistic opinion differences of different users at the same moment. It calculates the trust weights between users according to the user trust sets of each user. A mixed opinion evolution module is used to evolve the converted numerical opinion set and linguistic opinion set. During the evolution process, it calculates the numerical opinions and linguistic opinions at the next moment by using the numerical opinions and linguistic opinions at the previous moment and the trust weights between users. After the evolution is completed, a new mixed opinion set is formed. An iterative output module is used to iteratively execute the mixed opinion conversion, trust weight calculation, and mixed opinion evolution until the linguistic opinions and numerical opinions in the iteratively obtained mixed opinion set reach a consensus, and determines the recommended result output according to the current consensus state.

8. An electronic device, comprising a processor and a memory, the memory being used for storing a computer program, characterized in that, The processor is used to execute the computer program to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, The computer program, when executed by the processor, implements the method according to any one of claims 1 to 6.

Images