A fair recommendation method based on bias quantification and utility trade-off

Abstract

The application discloses a kind of fair recommendation methods based on bias quantification and utility trade-off, comprising: automatically classifying or manually classifying project, according to user group and project category, the bias value of a certain type of project is estimated by quantifying the rating of user to project to user group;With bias value, the rating of user is processed to remove bias, and input into recommendation algorithm to generate unbiased rating and generate unbiased predicted rating, take the top N project to generate recommendation list;Construct user attention model, select fairness index, construct the measurement strategy of utility and fairness;Exchange the position of recommended project in ranking list, calculate the change of measurement of ranking utility and fairness before and after exchange;Through exchange mechanism iteration updates recommendation list, generates the top-k recommendation list of optimal trade-off.The application realizes fair recommendation by data debiasing and utility trade-off, solves the data bias problem and the trade-off problem of fairness and utility of current recommendation system.

Description

Technical Field

[0001] This invention relates to the field of recommender system technology, and in particular to a fair recommender method based on bias quantification and utility trade-offs. Background Technology

[0002] Existing recommender systems often face fairness issues in optimizing user experience and commercial benefits, meaning that recommendation results may contain biases against certain user groups. These biases typically stem from historical biases in the training data or limitations of the model itself, leading to the systematic neglect or unfair treatment of certain user groups. Traditional fair recommendation methods usually focus only on eliminating bias, but this may reduce the utility of the recommendation. Therefore, how to maintain the utility of the recommender system while eliminating bias has become a critical problem that urgently needs to be solved. Summary of the Invention

[0003] To address the aforementioned shortcomings, this invention proposes a fair recommendation method based on bias quantification and utility trade-offs. This method achieves fair recommendations by removing data bias and weighing utility, thus resolving the data bias problem and the trade-off between fairness and utility in current recommendation systems.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a fair recommendation method based on bias quantification and utility trade-offs, comprising the following steps:

[0005] S1: Automatically categorize projects via online API services or manually categorize them via crowdsourcing services. Based on user group U and project category C, estimate the user group's bias value θ towards a certain type of project using the geometric mean of user u's ratings of project I. u ;

[0006] S2: Use bias values ​​to debias user ratings, generating unbiased ratings d. ui Input unbiased ratings into the recommendation algorithm to generate unbiased predicted ratings. Then, the top N items are selected to generate a recommendation list;

[0007] S3: Construct a user attention model, select appropriate fairness metrics, and construct measurement strategies for utility F and fairness G;

[0008] S4: Swap the positions of the recommended items in the ranking list, and calculate the changes in utility F and fairness G based on the rankings before and after the swap.

[0009] S5: Iteratively update the recommendation list through an exchange mechanism to generate the optimal top-k recommendation list.

[0010] As an improvement, the user group in step S1 includes the disadvantaged group D and the advantageous group A.

[0011] As an improvement, in step S1, for the disadvantaged group user U d For category c k Project Group The formula for the geometric mean of the scores is as follows:

[0012]

[0013] For the advantageous user group U a For category c k Project Group The formula for the geometric mean of the scores is as follows:

[0014]

[0015] Where, r ui This represents the rating that user u gives to item i.

[0016] As an improvement, in step S1, the bias value θ of the user group towards a certain type of project is... u The calculation formula is as follows:

[0017]

[0018] As an improvement, in step S2, for user u, the category is c k The formula for calculating the debiased score of item i is as follows:

[0019]

[0020] Where, r ui This represents the rating that user u gives to item i.

[0021] As an improvement, step S3 uses the inverse decay function to estimate the user's attention to construct a user attention model. The calculation formula is as follows:

[0022]

[0023] Among them, a m This represents the attention value at position m, α is a parameter that controls the decay rate; a larger α value means faster decay, and k is the number of the first k items.

[0024] As an improvement, the utility F in step S3 is calculated using the cumulative relevance score, and the calculation formula is as follows:

[0025]

[0026] In the above formula, a i It is the user's attention to item i. It is the predicted rating of user u for item i.

[0027] As an improvement, step S3 uses the Gini index as a fairness measure, and the calculation formula is as follows:

[0028]

[0029] In the above formula, E i It represents the cumulative exposure of the top i items in the recommended list.

[0030] As an improvement, the utility change ΔF after swapping the positions of items in step S4 is calculated using the following formula:

[0031] ΔF=F new -F old

[0032] In the above formula, F new F is the utility value after swapping the positions of the items. old This is the utility value before the items were swapped.

[0033] The fairness change ΔG after swapping project positions is calculated using the following formula:

[0034] ΔG=G new -G old

[0035] In the above formula, G new G is the fairness value after swapping the positions of the items. old It is the fairness value before the project positions are swapped.

[0036] As an improvement, the exchange mechanism in step S5 is to determine whether to accept the current exchange based on a preset threshold. If ΔF and ΔG satisfy... Then accept the exchange and update the current ranking. For projects after the recommended project position, recalculate the changes in utility F and fairness G until the desired result is achieved.

[0037] Compared with the prior art, the advantages of the present invention are as follows:

[0038] (1) This invention quantifies the bias value of user groups towards specific item categories and performs debiasing processing during the scoring process. This method significantly reduces the unfairness of recommendations caused by user bias, which helps to ensure that the recommendation system reflects the user's true interests and needs more objectively and fairly.

[0039] (2) By generating unbiased ratings and inputting them into the recommendation algorithm, this method can generate more accurate and reliable recommendation lists, which not only improves user satisfaction with the recommended content, but also enhances the overall performance and credibility of the recommendation system.

[0040] (3) When constructing the recommendation list, this method not only considers the recommendation utility (i.e., user satisfaction with the recommended items), but also takes fairness into account by constructing a user attention model and selecting appropriate fairness indicators. This dual consideration enables the recommendation results to meet the personalized needs of users and ensure fairness and diversity among different user groups.

[0041] (4) By iteratively updating the recommendation list and accepting or rejecting the exchange of project locations based on a preset trade-off threshold, this method provides high flexibility and adjustability, which allows the system to adjust the balance between fairness and utility according to actual needs to adapt to different application scenarios and user preferences.

[0042] (5) By continuously optimizing the order and combination of items in the recommendation list, this method can improve the user experience. Users can not only obtain recommended content that better matches their interests and needs, but also feel the importance and attention that the recommendation system attaches to fairness.

[0043] (6) Paying special attention to the needs of disadvantaged groups, this method helps to increase the diversity and inclusivity of recommended content. By reducing bias and discrimination, this method provides more equal and fair recommendation opportunities for different user groups. Attached Figure Description

[0044] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0045] Figure 1 This is a flowchart illustrating the overall process of the method of the present invention.

[0046] Figure 2 This is a flowchart of the bias quantification and debiasing recommendation module of the present invention;

[0047] Figure 3 This is a flowchart of the utility and fairness trade-off module of the present invention. Detailed Implementation

[0048] like Figures 1 to 3As shown, a fair recommendation method based on bias quantification and utility trade-offs includes a bias quantification module, a debiased recommendation module, and a utility and fairness trade-off module. The bias quantification module is used to predict the bias value of a user group. Specifically, it automatically categorizes items through online API services or manually through crowdsourcing services. Based on the user group and item category, it estimates the bias value of the user group towards a certain type of item using the geometric mean of user ratings. The debiased recommendation module is used to generate unbiased predicted ratings. Specifically, it uses the bias value to debias user ratings, generating unbiased ratings, which are then input into the recommendation algorithm to generate unbiased predicted ratings. The utility and fairness trade-off module is used to construct a user attention model. Specifically, it selects appropriate fairness indicators, constructs a utility and fairness measurement strategy, swaps the positions of recommended items in the ranking list, calculates the changes in utility and fairness measurements based on the rankings before and after the swap, and iteratively updates the recommendation list through the swap mechanism to obtain the optimally balanced recommendation list.

[0049] This invention is implemented in a recommendation scenario where the recommendation results are intended to balance utility and fairness. In this embodiment, a collaborative filtering algorithm is used to construct a recommendation model, and recommendations are made based on user-movie rating data from a movie recommendation platform.

[0050] In the following steps, U represents the user set, I represents the movie set, C represents the movie genre category (including action, suspense, period drama, romance, and comedy), A represents the user set associated with the advantageous group, and D represents the user set associated with the disadvantaged group. The advantageous and disadvantaged groups are divided according to the number of male and female users rating movies by genre (e.g., if more female users rate romance movies than male users, then female users belong to the advantageous group, and male users belong to the disadvantaged group). ui θ represents user u's rating of movie i. u This represents the bias value of user u.

[0051] A fair recommendation method based on bias quantification and utility trade-offs includes the following steps:

[0052] S1.1: Preprocess the data and classify the movies by genre through API services or crowdsourcing services.

[0053] S1.2: Quantify the bias, for disadvantaged user group U d For subject matter category c k user groups Geometric mean of the ratings:

[0054]

[0055] For the advantageous user group Ua For subject matter category c k user groups Geometric mean of the ratings:

[0056]

[0057] User u's bias value θ u Calculation formula:

[0058]

[0059] S2.1: Use bias values ​​to debias user ratings of movies, generating unbiased ratings.

[0060] For user u, the subject matter category is c k Movie i's debiased rating:

[0061]

[0062] Where, r ui θ represents user u's rating of movie i. u This represents the bias value of user u.

[0063] S2.2: Recommendation generation, which will generate the bias-free score d. ui Input collaborative filtering algorithm to generate predicted ratings for user u for movie i. Generate a candidate recommendation list by selecting the top N items;

[0064] S3: Measuring utility and fairness, utility F is calculated using cumulative predictive scores, and the formula is as follows:

[0065]

[0066] Among them, a i It is the user's attention to item i. It is a predicted score.

[0067] Fairness G is calculated using fairness measures such as the Gini index, and the formula is as follows:

[0068]

[0069] Among them, E i It represents the cumulative views, favorites, and plays of the top i movies in the recommended list.

[0070] S4: Starting from the initial ranking of the candidate recommendation list, adjust the ranking by swapping the positions of items i and j. Each swap will affect the utility and fairness metrics.

[0071] The change in utility ΔF after swapping i and j is calculated as follows:

[0072] ΔF=F new -F old

[0073] In the above formula, F new F is the utility value after swapping the positions of the items. old It is the utility value before the items were swapped.

[0074] The fairness change ΔG after swapping i and j is calculated as follows:

[0075] ΔG=G new -G old

[0076] In the above formula, G new G is the fairness value after swapping the positions of the items. old It is the fairness value before the project positions are swapped.

[0077] S5: Based on the preset threshold, determine whether to accept the current exchange. If ΔF and ΔG satisfy... If the exchange is accepted, the current ranking is updated, and step S6 is repeated until the target is reached. Finally, a top-k recommendation list with the optimal trade-offs is generated.

[0078] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0079] The units described in some embodiments of this disclosure can be implemented in software or in hardware. The described units can also be located in a processor, and the functions described above can be performed at least in part by one or more hardware logic components.

[0080] The present invention has been described above by way of example with reference to the accompanying drawings. Obviously, the specific implementation of the present invention is not limited to the above-described manner. Any non-substantial improvements made using the technical solution of the present invention, or the direct application of the concept and technical solution of the present invention to other occasions without modification, are all within the protection scope of the present invention.

Claims

1. A fair recommendation method based on bias quantification and utility trade-offs, characterized in that, Includes the following steps: S1: Automatically categorize projects via online API services or manually categorize them via crowdsourcing services. Based on user group U and project category C, estimate the user group's bias value for a certain type of project using the geometric mean of user u's ratings of project I. ; S2: Use bias values ​​to debias user ratings, generating unbiased ratings. Input unbiased ratings into the recommendation algorithm to generate unbiased predicted ratings. And take the top N items to generate a recommendation list; S3: Construct a user attention model, select appropriate fairness metrics, and construct measurement strategies for utility F and fairness G; S4: Swap the positions of the recommended items in the recommendation list, and calculate the changes in utility F and fairness G based on the rankings before and after the swap. S5: Iteratively update the recommendation list through an exchange mechanism to generate the top-k recommendation list with the best trade-offs; In step S2, for the user For category The formula for calculating the debiased score of item i is as follows: in, Indicates user For the project The rating; The change in utility ΔF after swapping the positions of items in step S4 is calculated using the following formula: In the above formula This is the utility value after swapping the positions of the items. This is the utility value before the items were swapped. The fairness change ΔG after swapping project positions is calculated using the following formula: In the above formula This is the fairness value after swapping the positions of the items. This is the fairness value before the items are swapped; In step S5, the exchange mechanism determines whether to accept the current exchange based on a preset threshold. and satisfy If the exchange is accepted and the current ranking is updated, the changes in utility F and fairness G for projects after the recommended project position are recalculated until the desired result is achieved. .

2. The fair recommendation method based on bias quantification and utility trade-off as described in claim 1, characterized in that: In step S1, the user groups include disadvantaged group D and advantageous group A.

3. The fair recommendation method based on bias quantification and utility trade-off as described in claim 2, characterized in that: In step S1, for disadvantaged groups of users For category Project Group The formula for the geometric mean of the scores is as follows: For advantageous user groups For category Project Group The formula for the geometric mean of the scores is as follows: in, Indicates user For the project Ratings This represents a set of items rated by disadvantaged users. This represents a set of items rated by a group of users with a strong user base.

4. The fair recommendation method based on bias quantification and utility trade-off as described in claim 3, characterized in that: In step S1, the bias value of the user group towards a certain type of project The calculation formula is as follows: 。 5. A fair recommendation method based on bias quantification and utility trade-offs according to claim 1, characterized in that: In step S3, the inverse decay function is used to estimate the user's attention to construct a user attention model. The calculation formula is as follows: in, Indicates the number in the recommended list Attention value at each position, It is a parameter that controls the decay rate, and a larger one... A higher value means faster decay. It was before The number of projects.

6. A fair recommendation method based on bias quantification and utility trade-offs according to claim 1, characterized in that: In step S3, utility F is calculated using the cumulative correlation score, and the calculation formula is as follows: In the above formula It is the user's opinion on the project attention, User For the project Predicted score This represents the number of recommended projects.

7. A fair recommendation method based on bias quantification and utility trade-offs according to claim 1, characterized in that: In step S3, the Gini index is used as a fairness measure, and the calculation formula is as follows: In the above formula The recommended list comes first. The cumulative exposure of each project Indicates the number in the recommended list One position.

Images