1. A recommendation method for a personalized product service scheme based on a trust level and a forgetting function is characterized by comprising the following steps:

calculating preference similarity according to historical preference data of the user;

fusing trust degree weight in the preference similarity to obtain final preference similarity;

calculating score similarity according to historical score data of the user;

fusing a time forgetting function in the scoring similarity to obtain a final scoring similarity;

fitting the final preference similarity and the final scoring similarity to obtain a combined similarity, and further searching Top-N neighbors of the user or the scheme;

and predicting the scores of the target users for the items according to the combined similarity, comparing the scores, and recommending a scheme for the target users.

2. The method for recommending personalized product service schemes based on confidence and forgetting functions according to claim 1, wherein the preference similarity comprises a preference similarity between users and a preference similarity between schemes, wherein the preference similarity between users is calculated by using a pearson correlation coefficient, and a specific calculation formula is as follows:

wherein p1 represents user U₁And user U₂Number of commonly used protocols, e_1,iRepresenting a user U₁Use scheme S_iThe number of times of the operation of the motor,representing a user U₁Average number of times of using all PSS schemes, e_2,iRepresenting a user U₂Use scheme S_iThe number of times of the operation of the motor,representing a user U₂Average number of times all PSS schemes are used;

the calculation formula of the preference similarity among the schemes is as follows:

wherein q1 denotes the simultaneous use of the PSS scheme S₁And scheme S₂Number of users, e_j,S1Representing a user U_jUse scheme S₁The number of times of the operation of the motor,represents scheme S₁Average number of times used by all users, e_j,S2Representing a user U_jUse scheme S₂The number of times of the operation of the motor,represents scheme S₂The average number of times used by all users.

3. The method for recommending a personalized product service scheme based on the credibility and forgetting function according to claim 2, wherein the credibility weight comprises the direct credibility between users and the indirect credibility between users, wherein the calculation formula of the direct credibility between users is as follows:

Trust_dir(U_a，U_b)＝Pre_Sim(U_a，U_b)*Jac(U_a，U_b)；

wherein, Trust_dir(U_a,U_b) Representing a user U_aFor user U_bDirect trust of, Pre _ Sim (U)_a，U_b) Representing a user U_aAnd user U_bPreference similarity of (1), Jac (U)_a,U_b) Representing a user U_aAnd user U_bJaccard similarity between, the user U_aAnd user U_bThe calculation formula of the Jaccard similarity between the two formulas is as follows:|I_a,bi represents the user U_aAnd user U_bNumber of commonly used scenarios, | I_aI represents the user U_aNumber of used solutions, | I_bI represents the user U_bThe number of used protocols;

the calculation formula of the indirect trust between the users is as follows:

wherein, Trust_dir(U_c,U_b) Representing a user U_cFor user U_bDirect Trust of, Trust_dir(U_a,U_c) Representing a user U_aFor user U_cDirect trust level of.

4. The method according to claim 3, wherein the final preference similarity includes a preference similarity between end users and a preference similarity between end solutions, wherein the preference similarity between end users is calculated by the following formula:

wherein, Trust (U)_a,U_b) Representing a user U_aFor user U_bIntegrated Trust of (U)_a,U_b)＝φTrust_dir(U_a,U_b)+(1-φ)Trust_ind(U_a,U_b) Phi denotes the degree of trust in the direct trust level between the users, Pre _ Sim (U)_a,U_b) Representing a user U_aAnd user U_bSimilarity of inter-preference;

the calculation formula of the preference similarity among the final schemes is as follows:

wherein Trust (S)₁,S₂) Represents scheme S₁To S₂Integrated Trust of (S), Trust (S)₁,S₂)＝φTrust_dir(S₁,S₂)+(1-φ)Trust_ind(S₁,S₂) Phi denotes the degree of confidence in the direct confidence between the schemes, Pre _ Sim (S)₁,S₂) Representing the similarity of preferences among the schemes.

5. The method for recommending personalized product service schemes based on confidence and forgetting functions according to claim 4, wherein the score similarity comprises inter-user score similarity and inter-scheme score similarity, wherein the inter-user score similarity is calculated by the following formula:

wherein p is²Representing a user U₁And user U₂Number of commonly scored schemes, w_1iRepresenting a user U₁For scheme S_iThe score of (a) is determined,representing a user U₁Average score for all PSS protocols, w_2iRepresenting a user U₂For scheme S_iThe score of (a) is determined,representing a user U₂Average scores for all PSS protocols;

the inter-scheme scoring similarity calculation formula is as follows:

wherein q is²Represents the scheme S for PSS₁And S₂Number of commonly scored schemes, w_j1Representing a user U_jFor scheme S₁The score of (a) is determined,represents scheme S₁Average score, w, scored by all users_j2Representing a user U_jFor scheme S₂The score of (a) is determined,represents scheme S₂Average score scored by all users.

6. The method for recommending a personalized product service scheme based on the credibility and forgetting function according to claim 5, wherein the calculation formula of the time forgetting function is as follows:

wherein alpha represents a forgetting coefficient, t represents the interval between the actual scoring time of the user and the reference time, and t represents the time interval between the actual scoring time of the user and the reference time_minRepresenting the interval, t, between the earliest scoring time of the user and the reference time_maxThe interval between the latest scoring time of the user and the reference time is represented.

7. The method according to claim 6, wherein the final scoring similarity includes a final inter-user scoring similarity and a final inter-solution scoring similarity, and the final inter-user scoring similarity is calculated by the following formula:

the calculation formula of the score similarity between the final schemes is as follows:

8. the method according to claim 7, wherein the combination similarity includes a combination similarity between users and a combination similarity between schemes; wherein, the calculation formula of the combination similarity among the users is as follows:

Com_Sim(U₁,U₂)＝β×Trust_Pre_Sim(U₁,U₂)+(1-β)Time_Sco_Sim(U₁,U₂)；

wherein beta represents a weight factor, beta is more than or equal to 0 and less than or equal to 1;

the calculation formula of the combination similarity among the schemes is as follows:

Com_Sim(S₁,S₂)＝γ×Trust_Pre_Sim(S₁,S₂)+(1-γ)Time_Sco_Sim(S₁,S₂)；

wherein gamma represents a weight factor, and gamma is more than or equal to 0 and less than or equal to 1.

Background

The product service system gradually becomes a main direction for transformation and upgrade of manufacturing enterprises, and research on the product service system gradually becomes a hot spot in the industry. At the same time, however, the manufacturing enterprise also faces the challenge of how to provide the personalized product service integration scheme to the manufacturing enterprise according to the user requirement, and some scholars study the modular design of the product service system to solve the problem.

Collaborative filtering is a common method for personalized recommendation technology, and has been widely applied to various recommendation systems. Although collaborative filtering techniques have met with great success in personalizing recommendations, two major problems remain: (1) cold start: the user or the scheme newly added into the platform has no historical use data, so that the similarity calculation cannot be carried out on the user or the scheme; (2) data sparsity: users typically score only a small portion of the solutions, resulting in two users or projects having no commonly scored items in the system and no similarity calculation. In view of the above method for solving the problems of cold start and data sparsity, the invention provides a mixed recommendation method based on the trust level and the forgetting function starting from two dimensions of the user and the project respectively.

Disclosure of Invention

In view of the above, the invention provides a recommendation method for a personalized product service scheme based on a trust level and a forgetting function, which introduces a trust level and a time forgetting weight in preference similarity calculation and scoring similarity calculation, and simultaneously comprehensively considers the combination similarity of users and schemes, so that the influence of data sparsity and cold start is reduced when a target user predicts a recommendation scheme, and the recommendation precision is improved.

In order to achieve the purpose, the invention adopts the following technical scheme: a recommendation method for a personalized product service scheme based on a trust level and a forgetting function specifically comprises the following steps:

calculating preference similarity according to historical preference data of the user;

fusing trust degree weight in the preference similarity to obtain final preference similarity;

calculating score similarity according to historical score data of the user;

fusing a time forgetting function in the scoring similarity to obtain a final scoring similarity;

fitting the final preference similarity and the final scoring similarity to obtain a combined similarity, and further searching Top-N neighbors of the user or the scheme;

and predicting the scores of the target users for the items according to the combined similarity, comparing the scores, and recommending a scheme for the target users.

Preferably, the preference similarity includes a preference similarity between users and a preference similarity between schemes, wherein the preference similarity between users is calculated by using a pearson correlation coefficient, and a specific calculation formula is as follows:

wherein p1 represents user U₁And user U₂Number of commonly used protocols, e_1,iRepresenting a user U₁Use scheme S_iThe number of times of the operation of the motor,representing a user U₁Average number of times of using all PSS schemes, e_2,iRepresenting a user U₂Use scheme S_iThe number of times of the operation of the motor,representing a user U₂Average number of times all PSS schemes are used;

the calculation formula of the preference similarity among the schemes is as follows:

wherein q1 denotes the simultaneous use of the PSS scheme S₁And scheme S₂Number of users, e_j,S1Representing a user U_jUse scheme S₁The number of times of the operation of the motor,represents scheme S₁Average number of times used by all users, e_j,S2Representing a user U_jUse scheme S₂The number of times of the operation of the motor,represents scheme S₂The average number of times used by all users.

Preferably, the trust level weight includes direct trust level between users and indirect trust level between users, wherein the calculation formula of the direct trust level between users is as follows:

Trust_dir(U_a，U_b)＝Pre_Sim(U_a，U_b)*Jac(U_a，U_b)；

wherein, Trust_dir(U_a,U_b) Representing a user U_aFor user U_bPre _ Sim represents the user U_aAnd user U_bPreference similarity of (1), Jac (U)_a,U_b) Representing a user U_aAnd user U_bJaccard similarity between, the user U_aAnd user U_bThe calculation formula of the Jaccard similarity between the two formulas is as follows:|I_a,bi represents the user U_aAnd user U_bNumber of commonly used scenarios, | I_aI represents the user U_aNumber of used solutions, | I_bI represents the user U_bThe number of used protocols;

the calculation formula of the indirect trust between the users is as follows:

wherein, Trust_dir(U_c,U_b) Representing a user U_bFor user U_cDirect trust level of.

By adopting the technical scheme, the method has the following beneficial technical effects: the traditional preference similarity calculation only considers the preference of users for the scheme, ignores the internal connection among the users, and has research that the users are more prone to accept the recommendation of acquaintances and friends than the recommendation of a system, and positive correlation exists between the trust of the users and the preference similarity. Moreover, the trust relationship between the users is merged into the recommendation system, and the accuracy is higher than that of the traditional collaborative filtering, so that the research on the mutual relationship between the users in the system is very necessary.

The direct trust between two users depends on whether a commonly used scheme exists between the two users, the direct trust is adopted only to easily cause inaccurate description of the users, and the commonly used scheme data between the users is less in practice, so that the direct trust between the users is less, and the problem of data sparsity can be greatly relieved by increasing the indirect trust.

Preferably, the final preference similarity includes a preference similarity between end users and a preference similarity between end solutions, wherein a calculation formula of the preference similarity between end users is as follows:

wherein, Trust (U)_a,U_b) Representing a user U_aTo U_bIntegrated Trust of (U)_a,U_b)＝φTrust_dir(U_a,U_b)+(1-φ)Trust_ind(U_a,U_b) Phi denotes the degree of trust in the direct trust level between the users, Pre _ Sim (U)_a,U_b) Representing preference similarity among the users;

the calculation formula of the preference similarity among the final schemes is as follows:

wherein Trust (S)₁,S₂) Represents scheme S₁To S₂Integrated Trust of (S), Trust (S)₁,S₂)＝φTrust_dir(S₁,S₂)+(1-φ)Trust_ind(S₁,S₂) Phi denotes the trust of the direct trust level between the schemesDegree, Pre _ Sim (S)₁,S₂) Representing the similarity of preferences among the schemes.

Preferably, the score similarity includes a score similarity between users and a score similarity between schemes, wherein a calculation formula of the score similarity between users is as follows:

wherein p is²Representing a user U₁And user U₂Number of commonly scored schemes, w_1iRepresenting a user U₁For scheme S_iThe score of (a) is determined,representing a user U₁Average score for all PSS protocols, w_2iRepresenting a user U₂For scheme S_iThe score of (a) is determined,representing a user U₂Average scores for all PSS protocols;

the inter-scheme scoring similarity calculation formula is as follows:

wherein q is²Represents the scheme S for PSS₁And S₂Number of commonly scored schemes, w_j1Representing a user U_jFor scheme S₁The score of (a) is determined,represents scheme S₁Average score, w, scored by all users_j2Representing a user U_jFor scheme S₂The score of (a) is determined,represents scheme S₂Is evaluated by all usersThe average score of the scores.

Preferably, the calculation formula of the time forgetting function is as follows:

wherein alpha represents a forgetting coefficient, t represents the interval between the actual scoring time of the user and the reference time, and t represents the time interval between the actual scoring time of the user and the reference time_minRepresenting the interval, t, between the earliest scoring time of the user and the reference time_maxThe interval between the latest scoring time of the user and the reference time is represented.

By adopting the technical scheme, the method has the following beneficial technical effects: the traditional score similarity calculation gives the scores of the users at different time with the same weight, and does not consider the actual situation that the interests of the users change along with the change of the time, so the influence of the time is necessary to be considered when the score similarity is calculated, and the interests of the users also change along with the change of the time according to the process of forgetting by human, so the score similarity calculation also conforms to the rule of forgetting curves.

Preferably, the final scoring similarity includes a final inter-user scoring similarity and a final inter-scheme scoring similarity, wherein a calculation formula of the final inter-user scoring similarity is as follows:

the calculation formula of the score similarity between the final schemes is as follows:

preferably, the combination similarity includes a combination similarity between users and a combination similarity between schemes; wherein, the calculation formula of the combination similarity among the users is as follows:

Com_Sim(U₁,U₂)＝β×Trust_Pre_Sim(U₁,U₂)+(1-β)Time_Sco_Sim(U₁,U₂)；

wherein beta represents a weight factor, beta is more than or equal to 0 and less than or equal to 1;

the calculation formula of the combination similarity among the schemes is as follows:

Com_Sim(S₁,S₂)＝γ×Trust_Pre_Sim(S₁,S₂)+(1-γ)Time_Sco_Sim(S₁,S₂)；

wherein gamma represents a weight factor, and gamma is more than or equal to 0 and less than or equal to 1.

By adopting the technical scheme, the method has the following beneficial technical effects: the preference similarity among the end users and the scoring similarity among the end users, and the preference similarity among the end schemes and the scoring similarity among the end schemes are combined, and even under the condition of lacking a certain similarity, the combination similarity among the users or the combination similarity among the schemes can be obtained, so that the problems of data sparsity and cold start are relieved to a certain extent.

According to the technical scheme, on the basis of a traditional collaborative filtering algorithm, a recommendation method of a personalized product service system based on trust and time forgetting is provided, compared with the prior art, trust and time forgetting weight are introduced in preference similarity calculation and scoring similarity calculation, and meanwhile, the combined similarity of users and schemes is comprehensively considered, so that the influence of data sparsity and cold start can be reduced when a target user is predicted to recommend the scheme, and the recommendation precision is improved.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.

FIG. 1 is a flow chart of the algorithm of the present invention;

FIG. 2 is a graph showing the comparison of the algorithm of the present invention with other algorithms.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

The embodiment of the invention discloses a recommendation method of a personalized product service scheme based on a trust degree and a forgetting function, which comprises the following steps as shown in figure 1:

the method comprises the following steps: calculating preference similarity among users and preference similarity among schemes according to historical preference data of the users;

it should be noted that the preference similarity between users refers to the similarity of users when selecting schemes, and the greater the preference similarity between users, the more similar the schemes selected by the two users are; the inter-scheme preference similarity refers to the similarity degree of the two schemes on the audiences, and the larger the inter-scheme preference similarity is, the more similar the audience groups of the two schemes are.

Specifically, in this embodiment, the bias similarity between users is calculated by using the pearson correlation coefficient, and the specific calculation formula is as follows:

wherein p1 represents user U₁And user U₂Number of commonly used protocols, e_1,iRepresenting a user U₁Use scheme S_iThe number of times of the operation of the motor,representing a user U₁Using all PSS schemesAverage number of times, e_2,iRepresenting a user U₂Use scheme S_iThe number of times of the operation of the motor,representing a user U₂Average number of times of using all PSS schemes.

Pre_Sim(U₁,U₂) Is in the range of [ -1, 1 [)]When the value is-1 or 1, U₁And U₂Is completely correlated, when the value is 0, U₁And U₂If p1 is 0, Pre _ Sim (U)₁,U₂) Is null.

Similarly, the calculation formula of the preference similarity among the schemes is as follows:

wherein q1 denotes the simultaneous use of the PSS scheme S₁And S₂Number of users, e_j,S1Representing a user U_jUse scheme S₁The number of times of the operation of the motor,represents scheme S₁Average number of times used by all users, e_j,S2Representing a user U_jUse scheme S₂The number of times of the operation of the motor,represents scheme S₂The average number of times used by all users.

Pre_Sim(S₁,S₂) Is in the range of [ -1, 1 [)]When the value is-1 or 1, S₁And S₂Is completely correlated, when the value is 0, S₁And S₂Is completely irrelevant if q1 is 0, Pre _ Sim (U)₁,U₂) Is null.

Step two: fusing the trust degree weight in the preference similarity between the users and the preference similarity between the schemes to obtain the preference similarity between the final users and the preference similarity between the final schemes;

the traditional preference similarity calculation only considers the preference of users for the scheme, ignores the internal connection among the users, and has research that the users are more prone to accept the recommendation of acquaintances and friends than the recommendation of a system, and positive correlation exists between the trust of the users and the preference similarity. Moreover, the trust relationship between the users is merged into the recommendation system, and the accuracy is higher than that of the traditional collaborative filtering, so that the research on the mutual relationship between the users in the system is very necessary.

The trust degree weight comprises direct trust degree between users and indirect trust degree between users.

Further, the calculation formula of the direct trust between users is as follows:

Trust_dir(U_a，U_b)＝Pre_Sim(U_a，U_b)*Jac(U_a，U_b)；

wherein, Trust_dir(U_a,U_b) Representing a user U_aTo U_bPre _ Sim represents the user U_aAnd U_bPreference similarity of (1), Jac (U)_a,U_b) Representing a user U_aAnd U_bJaccard similarity between.

User U_aAnd U_bThe calculation formula of the Jaccard similarity between the two formulas is as follows:

wherein, | I_a,bI represents the user U_aAnd user U_bNumber of commonly used scenarios, | I_aI represents the user U_aNumber of used solutions, | I_bI represents the user U_bNumber of used protocols. The Jaccard similarity value range is [0, 1 ]]The larger the Jaccard similarity is, the higher the relevance of the user is.

The direct trust between two users depends on whether a commonly used scheme exists between the two users, the direct trust is adopted only to easily cause inaccurate description of the users, and the commonly used scheme data between the users is less in practice, so that the direct trust between the users is less, and the problem of data sparsity can be greatly relieved by increasing the indirect trust.

Further, the calculation formula of the indirect trust between users is as follows:

by combining the direct trust and the indirect trust between the users, the calculation formula of the comprehensive trust between the users is as follows:

Trust(U_a,U_b)＝φTrust_dir(U_a,U_b)+(1-φ)Trust_ind(U_a,U_b)；

wherein phi represents the trust degree of the direct trust degree between the users, and the larger phi represents the larger contribution of the direct trust degree to the comprehensive trust degree.

Further, the calculation formula of the preference similarity between the end users is as follows:

similarly, the calculation formula of the preference similarity between the final schemes is as follows:

step three: calculating the scoring similarity between users and the scoring similarity between schemes according to the historical scoring data of the users;

the calculation formula of the scoring similarity among the users is as follows:

wherein p2 represents user U₁And user U₂Number of commonly scored schemes, w_1iRepresenting a user U₁For scheme S_iThe score of (a) is determined,representing a user U₁Average score for all PSS protocols, w_2iRepresenting a user U₂For scheme S_iThe score of (a) is determined,representing a user U₂Average scores for all PSS schemes.

Similarly, the calculation formula of the score similarity between the schemes is as follows:

wherein q2 denotes the scheme S for PSS₁And S₂Number of commonly scored schemes, w_j1Representing a user U_jFor scheme S₁The score of (a) is determined,represents scheme S₁Average score, w, scored by all users_j2Representing a user U_jFor scheme S₂The score of (a) is determined,represents scheme S₂Average score scored by all users.

Step four: fusing a time forgetting function in the inter-user scoring similarity and the inter-scheme scoring similarity to obtain final inter-user scoring similarity and final inter-scheme scoring similarity;

the traditional score similarity calculation gives the scores of the users at different times the same weight, and does not consider the actual situation that the interests of the users change along with the change of the time, so that the influence of the time is necessarily considered when the score similarity is calculated. According to the process of forgetting by human, the interest of the user can change along with the change of time, so that the method also conforms to the rule of forgetting curves.

Setting the interval between the actual scoring time and the reference time of the user as t, and setting the interval between the earliest scoring time and the reference time of the user as t_min＝min(t_a-t_r) The earliest scoring time refers to the time when the user scores the scheme for the first time, and the interval t between the latest scoring time of the user and the reference time_max＝max(t_a-t_r) Wherein t is_rFor reference time, t_aThe actual time of scoring the plan for the user. In this embodiment, a time forgetting function is introduced during score similarity calculation, and the calculation formula is as follows:

wherein alpha represents a forgetting coefficient, and the forgetting coefficient is in a direct proportion relation with the interest change among users. When alpha is 1, the preference of the user is changed, when 0 < alpha < 1, the user is partially forgotten nonlinearly, and when alpha is 0, the preference of the user is not changed.

The calculation formula of the scoring similarity between the end users with the time forgetting function fused is as follows:

similarly, the calculation formula of the score similarity between the final schemes fused with the time forgetting function is as follows:

step five: fitting the preference similarity between the end users and the scoring similarity between the end users to obtain the combined similarity between the users, fitting the preference similarity between the final schemes and the scoring similarity between the final schemes to obtain the combined similarity between the schemes, and further searching the Top-N neighbor of the user or the scheme;

user U₁And user U₂Combined similarity ofThe calculation formula is as follows:

Com_Sim(U₁,U₂)＝β×Trust_Pre_Sim(U₁,U₂)+(1-β)Time_Sco_Sim(U₁,U₂)；

wherein beta represents a weight factor, and beta is more than or equal to 0 and less than or equal to 1.β ═ 0 means that there is only score similarity, β ═ 1 means that there is only preference similarity, and for user U, there is only preference similarity_iAnd calculating the similarity between the neighbor cell and other users and filtering out users smaller than or equal to a preset threshold value to obtain the neighbor based on the user.

In a similar manner, the PSS scheme S₁And S₂The combined similarity calculation formula is as follows:

Com_Sim(S₁,S₂)＝γ×Trust_Pre_Sim(S₁,S₂)+(1-γ)Time_Sco_Sim(S₁,S₂)；

wherein gamma represents a weight factor, and gamma is more than or equal to 0 and less than or equal to 1.γ ═ 0 indicates that only score similarity exists, and γ ═ 1 indicates that only preference similarity exists. For scheme S₁And calculating the similarity between the scheme and other schemes and filtering out the schemes with the similarity being less than or equal to a preset threshold value, so as to obtain the neighbors based on the scheme.

According to the embodiment of the invention, the preference similarity between the end users and the score similarity between the end users, and the preference similarity between the end schemes and the score similarity between the end schemes are combined, and even if a certain similarity is lacked, the combination similarity between the users or the combination similarity between the schemes can be obtained, so that the problems of data sparsity and cold start are relieved to a certain extent.

Step six: and predicting the scores of the target users for the items according to the combination similarity between the users and the combination similarity between the schemes, and comparing the scores to recommend the schemes for the target users.

The filtering method based on the user is to use the following formula to carry out scoring prediction after finding the nearest neighbor of the user:

wherein, R (U)_i,jRepresenting a user U_iFor scheme S_jThe prediction score of (a) is determined,representing a user U_iAverage score for all schemes, Com _ Sim (U)_i,U_k) Representing a user U_iAnd U_kCombined similarity between, R_k,jRepresenting a user U_kFor scheme S_jThe score of (a) is determined,representing a user U_kIs in contact with U_iAverage score in a common scoring schema.

The scheme-based filtering method is to perform score prediction by using the following formula after finding the nearest neighbor of a scheme:

wherein, R (S)_i,jRepresenting a user U_iFor scheme S_jThe prediction score of (a) is determined,represents a user pair S_jAverage score of all scores, Com _ Sim (S)_i,S₁) Represents scheme S_jAnd S₁Combined similarity between, w_i,lRepresenting a user U_iFor scheme S₁The score of (a) is determined,represents scheme S₁At with S_jThe average score on the user is scored jointly.

In the prediction of the rating, when there is no similar user who scores the target plan, a plan-based filtering method is employed, and when there is no similar plan scored by the target user, a user-based filtering method is employed. If both can be used, a combined filtering method is adopted, and the formula is as follows:

example (c): the analysis data of the embodiment is from a certain sensor manufacturing company in Dandong, and the enterprise professionally produces the geotechnical engineering measurement steel string type series sensor, and proposes recommendation for providing a personalized scheme for the user in order to improve the sales capability of the enterprise and meet the personalized requirements of consumers. By sorting the past user purchase records, 10 representative schemes are selected for analysis in the embodiment, and the 10 schemes are respectively a steel bar stress meter series (S)₁) Surface strain gauge series (S)₂) Surface stress meter series (S)₃) Pressure cell series (S)₄) Load meter series (S)₅) Displacement meter series (S)₆) Temperature measuring instrument series (S)₇) Anchor dynamometer series (S)₈) Axial force meter series (S)₉) Harmonic vibration frequency detector series (S)₁₀). According to the sales records of the enterprise, historical data of 10 users are selected for carrying out analysis of the personalized recommendation process, wherein the historical data comprises the use times, scores and scoring time of the schemes of the 10 users. The score values are integers between 1 and 5, with 1 to 5 indicating very unsatisfactory, less satisfactory, general, more satisfactory and very satisfactory, respectively. By user U₁For example, the user has used scheme S₁、S₄、S₆Now from scheme S₂、S₃、S₅、S₇、S₈、S₉、S₁₀User U in₁Recommending the most appropriate scheme, wherein the recommendation process comprises the following steps:

1. according to the preference matrix of the user-scheme, the original preference similarity between users and the preference similarity between schemes are calculated, and the corresponding preference matrix is shown in table 5 and table 6.

TABLE 5 similarity of preferences among users

TABLE 6 preference similarity between schemes

2. According to the user-scheme preference matrix, the trust between users and schemes is calculated, the direct trust coefficient phi is 0.8, and then the preference similarity matrix based on the trust can be obtained by calculation with a formula, as shown in tables 7 and 8.

TABLE 7 inter-user preference similarity with fusion confidence

TABLE 8 inter-schema preference similarity with fusion of confidence

3. According to the user scoring matrix, the time forgetting coefficient α is also taken to be 0.5, and a scoring similarity matrix based on the time forgetting weight is calculated by using a formula, as shown in tables 9 and 10.

TABLE 9 inter-user score similarity fusing time forget function

TABLE 10 inter-scheme score similarity fusing time forget function

4. The preference similarity and the score similarity between the users and the schemes are combined, and a combined similarity matrix between the users and the schemes is calculated by using a formula, as shown in tables 11 and 12.

TABLE 11 inter-user combination similarity

TABLE 12 combination similarity between schemes

5. In the embodiment, the nearest neighbor number of the user is 3, and U can be obtained according to the combination similarity between the users₁Three users with the highest similarity are respectively U₅、U₁₀、U₈. Calculating a prediction U₁For unused scheme S₂、S₃、S₅、S₇、S₈、S₉、S₁₀The scores of (a) are shown in Table 13. The first three schemes with the highest predictive score are S₃、S₇And S₅And is thus the target user U₁Recommendation scheme S₃、S₇And S₅。

TABLE 13 prediction scoring matrix

To verify the effectiveness of the recommended algorithm, the average absolute error is used as an evaluation criterion, and the algorithm is compared with other collaborative filtering algorithms. The Mean Absolute Error (MAE) is often used to evaluate the performance of the recommender system, and the accuracy of the recommender system is measured by calculating the mean absolute deviation between the predicted score of the recommender system and the actual score of the user. The smaller the mean absolute error value, the higher the accuracy of the recommendation system. Let R ' ═ R ' be the actual scoring set of the user '₁，R′₂，…，R′_nThe system predicts the evaluation diversity as R ═ R { (R)₁，R₂，…，R_nThe calculation formula of MAE is:

the confidence coefficient phi is set to 0.8, the forgetting coefficient alpha is set to 0.2, and the scheme neighbor number L is set to 5, and the algorithm is compared with the following algorithms, as shown in fig. 2.

As can be seen from fig. 2, compared with other algorithms, the recommendation algorithm proposed herein has a smaller MAE value under different number of neighbor sets, and has a good recommendation effect. Therefore, when the similarity is calculated, the change of the interest of the user can be more accurately mastered by considering the trust and the time forgetting weight, and the prediction accuracy can be obviously improved.

The invention provides a recommendation algorithm of a personalized product service system based on trust and time forgetting on the basis of a traditional collaborative filtering algorithm. And respectively calculating preference similarity and score similarity according to the historical records of the users, introducing a trust factor when calculating the preference similarity, and introducing a time factor when calculating the score similarity. In addition, the accuracy of the score prediction is improved by further calculating the combination similarity between users and schemes, and the method is favorable for finding the optimal recommendation scheme. Through the analysis of the personalized recommendation process of a certain sensor enterprise and the comparison with the traditional collaborative filtering algorithm based on the user, the algorithm has higher precision and better recommendation effect than other algorithms according to the experimental result.

On the basis of the previous research, the invention introduces trust factors and time factors, more comprehensively considers the internal relationship of the system and makes up the defects caused by only considering a certain factor. Meanwhile, the invention considers the combination similarity between users and the combination similarity between schemes, so that the score prediction can be carried out even if a certain similarity is lacked, and the problems of data sparsity and cold start are greatly relieved.

The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. The device disclosed by the embodiment corresponds to the method disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the method part for description.

The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.