Recommendation system article side individual fairness optimization method based on activation function

Abstract

The invention discloses a recommendation system article side individual fairness optimization method based on an activation function, and the method can carry out the article side individual fairness optimization based on a recommendation strategy given by a basic recommendation model, and specifically comprises the IED micro progressive optimization based on the activation function and the optimization based on random sampling. The object-side individual fairness optimization item can be used as the regularization loss of the base model, and the fairness optimization of the final recommendation strategy is carried out. According to the item individual side fairness optimization method of the recommendation system based on the activation function provided by the invention, the item individual side fairness of the recommendation system can be optimized so as to relieve the Mortai effect.

Description

Technical Field

[0001] The present invention relates to the technical field of recommendation algorithm application, and in particular to an activation function-based recommendation system item-side individual fairness optimization method. Background Art

[0002] Recommendation systems play a vital role in today's information age. They not only help users discover new content and products, but also promote the activity and profitability of online platforms. By analyzing user preferences and behaviors, recommendation systems can provide personalized recommendations, improve user experience, increase user stickiness, and facilitate transaction completion. The fairness of recommendation systems is also crucial. A fair recommendation system can not only increase users' trust in recommendation algorithms, but also reduce information filtering and bias, ensuring that users receive diverse information and product recommendations. Fairness also helps maintain the reputation of the platform, attract more users and merchants to participate, and promote the healthy development of the entire ecosystem.

[0003] However, in current recommendation systems, item-side fairness is often ignored, resulting in some items being over-recommended while others being ignored. This unfairness will trigger the Matthew effect, making popular items even more popular and unpopular items even less popular, exacerbating market imbalances. This unfairness will not only affect the diversity and satisfaction of users’ choices, but will also affect merchants’ profits and item exposure, ultimately affecting the effectiveness and sustainability of the entire recommendation system. Therefore, when it comes to algorithms, it is necessary to consider the optimization of item-side fairness in order to solve the unfairness issues that exist in current recommendation systems. Summary of the invention

[0004] The purpose of the present invention is to provide a method for optimizing the fairness of individual items in a recommendation system based on an activation function in view of the deficiencies in the prior art, so that the fairness of items in the recommendation system can be optimized to alleviate the Matthew effect.

[0005] The objective of the present invention is achieved through the following technical solution: a method for optimizing individual fairness of items in a recommendation system based on an activation function, comprising the following steps:

[0006] S1. Obtain the recommendation strategy of the base model for the user set;

[0007] S2, a progressive method based on the ranking function of the activation function, which obtains the ranking of items in the recommendation strategy through differentiable terms;

[0008] S3, calculate the average exposure of items in the user set by ranking the items in the recommendation strategy;

[0009] S4, calculating the exposure difference index according to the average exposure of each item and the sum of the absolute values ​​of the exposure differences between all item pairs in the recommendation strategy;

[0010] S5. Use the exposure difference index as the loss function to optimize the base model, and use the optimized base model to obtain the final optimized recommendation strategy.

[0011] Furthermore, the recommendation strategy is a collection of top-k recommendation strategies of n users for m items.

[0012] Furthermore, the progressive method of the ranking function of the activation function is specifically:

[0013] Define the score of item i as r i Then there is

[0014]

[0015] in This is a top-k recommendation strategy for a single user, and the calculation process is all differentiable.

[0016] Furthermore, the calculation of the average exposure of the item in the user set includes:

[0017] Item i in the recommendation strategy The exposure level in is expressed as:

[0018]

[0019] Average exposure of items in user collections:

[0020]

[0021] Where U represents the user set.

[0022] Furthermore, the calculation of the exposure difference index includes:

[0023]

[0024] This value can be used to measure individual fairness on the item side. It represents the sum of the absolute values ​​of the exposure differences between all pairs of items in the recommendation strategy, reflecting the degree of exposure imbalance between items. is a normalization factor to make the exposure differences between different recommended strategies comparable.

[0025] Furthermore, in the calculation of the exposure difference index, random partial sampling is used to reduce the time complexity:

[0026]

[0027]

[0028] According to another aspect of the specification, there is also provided a device for optimizing individual fairness of items in a recommendation system based on an activation function, comprising a memory and one or more processors, wherein executable code is stored in the memory, and when the processor executes the executable code, the method for optimizing individual fairness of items in a recommendation system based on an activation function is implemented.

[0029] According to another aspect of the specification, a computer-readable storage medium is also provided, on which a program is stored. When the program is executed by a processor, the method for optimizing individual fairness on the item side of a recommendation system based on an activation function is implemented.

[0030] Beneficial effects of the present invention:

[0031] Aiming at the non-differentiable problem in the IED calculation process, the present invention adopts the problem of approximating the ranking function based on the activation function, solves the traditional IED unoptimizable process, improves the progressive process, saves the calculation cost, and realizes the optimization of individual fairness on the item side of the recommendation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 Flowchart of the method for optimizing individual fairness on the item side of the recommendation system based on activation function.

[0033] Figure 2 Calculation process diagram for IED.

[0034] Figure 3 This is a graph showing the specific implementation results of this method after optimization on the Amazon CDs and Vinyl dataset.

[0035] Figure 4 Schematic diagram of the device for optimizing fairness of individual items in a recommendation system based on activation function. DETAILED DESCRIPTION

[0036] The specific implementation modes of the present invention are further described in detail below with reference to the accompanying drawings.

[0037] like Figure 1 As shown, the present invention provides a method for optimizing individual fairness of items in a recommendation system based on an activation function, comprising the following steps:

[0038] Step 1: Get the recommended strategy of the base model. represents the top-k recommendation strategy of the recommendation model for the user set. Specifically, there is a user set U = {u1,u2,,u n} and the item set I = {i1,i2,,im}. Recommended strategy is a collection of top-k recommendation strategies of multiple users. It represents the top-k recommendation strategy for a single user. Based on the ratings of all items in the item set I given by user u in the basic recommendation model, the top k items are selected as recommended items. A collection of items.

[0039] Step 2: Optimize the individual fairness of items. First, determine the definition of fairness on the individual side of items. Then, according to its non-differentiable nature and computational complexity, the IED differentiable progressive scheme based on activation function and the optimization based on random sampling are adopted respectively.

[0040] This step is the core step of the present invention. Figure 2 As shown, it is divided into the following sub-steps:

[0041] (2.1) Definition of individual fairness on the item side: First, define the position deviation of item i.

[0042]

[0043] Indicates item i in the recommendation strategy The higher the item i is in the recommendation strategy, the greater the exposure value it gets.

[0044] Then, the total exposure of item i is defined as the average exposure obtained by item i in the recommendation strategy for all user sets.

[0045]

[0046] Different items have different exposure values, which creates individual unfairness on the item side. Therefore, the item exposure disparity index (IED) is defined as

[0047]

[0048] This value can be used to measure individual fairness on the item side. This part represents the sum of the absolute values ​​of the exposure differences between all item pairs in the recommendation strategy, which reflects the degree of exposure imbalance between items. is a normalization factor, which ensures that the value of IED is within a reasonable range, making the exposure differences between different recommendation strategies comparable. Where |I| represents the size of the item set I, that is, the total number of items, and i, i', i" are all subset elements in the item set.

[0049] (2.2) IED Differentiable Progressive Scheme Based on Activation Function

[0050] Based on the item-side individual fairness evaluation index IED given in (2.1), due to the existence of a rank process, that is, Function, to obtain the ranking of item i in the recommendation strategy, such an operation is not differentiable. That is, the IED indicator can only be used to evaluate the individual fairness of the item side and cannot be directly used to optimize the individual fairness of the item side. The problem lies in the non-differentiable nature of the rank function.

[0051] Using the progressive method of the ranking function based on the activation function, the score of item i is defined as r i Then there is

[0052]

[0053] The calculation process is all differentiable, so the final IED can be differentiable.

[0054] (2.3) Optimization based on random sampling

[0055] The calculation of IED in (2.1) requires calculating the total number of items in the item set I. However, the experiment found that due to the association between items, partial random sampling can also achieve similar effects, and a sampling method of approximately 1 / 4 items can be used.

[0056]

[0057] The method based on random partial sampling can effectively reduce the computational time complexity without causing excessive loss in the computational results.

[0058] Step 3: The exposure difference index is the individual fairness index on the item side, which is used as the loss for base model optimization. Specifically, the exposure difference index obtained in step 2 after activation function asymptotic optimization and random sampling optimization can be used as the loss of the base model, and the fairness of the item side can be effectively optimized through gradient calculation and parameter update of the base model parameters.

[0059] The loss function L of the base model can be defined as

[0060]

[0061] in

[0062]

[0063] The base model can be any model that can output a recommendation strategy π k UA deep learning model can be used to obtain a fairer recommendation strategy by optimizing the loss function L for multiple rounds, thereby reducing the IED value. The optimization process is to calculate the gradient of the base model and update the parameters through the optimizer.

[0064] The specific implementation results of the optimization scheme for individual fairness of recommendation system items based on activation function on the Amazon CDs and Vinyl dataset are as follows: Figure 3 shown.

[0065] As the number of training rounds increases, the value of IED decreases effectively, which indicates that individual fairness is optimized. The optimization scheme of individual fairness on the item side of the recommendation system based on activation function is feasible and effective.

[0066] Corresponding to the aforementioned embodiment of a method for optimizing individual fairness of items in a recommendation system based on an activation function, the present invention also provides an embodiment of a device for optimizing individual fairness of items in a recommendation system based on an activation function.

[0067] See also Figure 4 An embodiment of the present invention provides an individual fairness optimization device for an item in a recommendation system based on an activation function, comprising a memory and one or more processors, wherein the memory stores executable code, and when the processor executes the executable code, it is used to implement an individual fairness optimization method for an item in a recommendation system based on an activation function in the above embodiment.

[0068] The embodiment of the device for optimizing individual fairness on the item side of a recommendation system based on an activation function provided by the present invention can be applied to any device with data processing capabilities, and the device with data processing capabilities can be a device or apparatus such as a computer. The device embodiment can be implemented through software, or through hardware or a combination of software and hardware. Taking software implementation as an example, as a device in a logical sense, it is formed by the processor of any device with data processing capabilities in which it is located reading the corresponding computer program instructions in the non-volatile memory into the internal memory for execution. From a hardware perspective, if Figure 4 As shown, it is a hardware structure diagram of any device with data processing capability where the recommendation system item-side individual fairness optimization device based on activation function provided by the present invention is located, except Figure 4 In addition to the processor, memory, network interface, and non-volatile memory shown, any device with data processing capabilities in which the apparatus in the embodiments is located may also include other hardware, generally based on the actual functions of the device with data processing capabilities, which will not be described in detail.

[0069] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0070] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of the present invention. Ordinary technicians in this field can understand and implement it without paying creative work.

[0071] An embodiment of the present invention further provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, a method for optimizing individual fairness on the item side of a recommendation system based on an activation function in the above embodiment is implemented.

[0072] The computer-readable storage medium may be an internal storage unit of any device with data processing capability described in any of the aforementioned embodiments, such as a hard disk or a memory. The computer-readable storage medium may also be an external storage device of any device with data processing capability, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc. equipped on the device. Furthermore, the computer-readable storage medium may also include both an internal storage unit and an external storage device of any device with data processing capability. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capability, and may also be used to temporarily store data that has been output or is to be output.

[0073] The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements the method for optimizing individual fairness of items in a recommendation system based on an activation function.

[0074] Those skilled in the art will readily appreciate other embodiments of the present application after considering the description and practicing the contents disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The description and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the claims.

[0075] It should be understood that the above general description and the detailed description below are only exemplary and explanatory and cannot limit the present application. The present application is not limited to the precise structure described above and shown in the drawings, and various modifications and changes can be made without departing from the scope thereof. The scope of the present application is limited only by the attached claims.

Claims

1. A method for optimizing individual fairness of items in a recommendation system based on activation function, characterized in that: The method includes: S1. Obtain the recommendation strategy of the base model for the user set; S2, a progressive method based on the ranking function of the activation function, which obtains the ranking of items in the recommendation strategy through differentiable terms; S3, calculating the average exposure of items in the user set by ranking the items in the recommendation strategy; S4, calculating the exposure difference index according to the average exposure of each item and the sum of the absolute values ​​of the exposure differences between all item pairs in the recommendation strategy; S5. Use the exposure difference index as the loss function to optimize the base model, and use the optimized base model to obtain the final optimized recommendation strategy.

2. According to claim 1, the method for optimizing individual fairness of items in a recommendation system based on activation function is characterized in that: The recommendation strategy is a collection of top-k recommendation strategies of n users for m items.

3. According to claim 1, the method for optimizing individual fairness of items in a recommendation system based on activation function is characterized in that: The progressive method of the sorting function of the activation function is specifically: Define the score of item i as r i Then there is in This is a top-k recommendation strategy for a single user, and the calculation process is all differentiable.

4. According to claim 3, the method for optimizing individual fairness of items in a recommendation system based on activation function is characterized in that: The calculation of the average exposure of the item in the user set includes: Item i in the recommendation strategy The exposure level in is expressed as: Average exposure of items in user collections: Where U represents the user set.

5. According to claim 1, the method for optimizing individual fairness of items in a recommendation system based on activation function is characterized in that: The calculation of the exposure difference index includes: This value can be used to measure individual fairness on the item side. It represents the sum of the absolute values ​​of the exposure differences between all pairs of items in the recommendation strategy, reflecting the degree of exposure imbalance between items. is a normalization factor to make the exposure differences between different recommended strategies comparable.

6. According to claim 5, the method for optimizing individual fairness of items in a recommendation system based on activation function is characterized in that: In the calculation of the exposure difference index, random partial sampling is used to reduce the time complexity: Where |I| represents the size of the item set I, that is, the total number of items, and i, i', i" are all subset elements in the item set.

7. A device for optimizing individual fairness of items in a recommendation system based on an activation function, comprising a memory and one or more processors, wherein executable code is stored in the memory, characterized in that: When the processor executes the executable code, an individual fairness optimization method for item side of a recommendation system based on an activation function is implemented as described in any one of claims 1 to 6.

8. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, an individual fairness optimization method for item side of a recommendation system based on an activation function is implemented as described in any one of claims 1 to 6.

Images