An advertisement fusion recommendation algorithm based on operational optimization

Abstract

This invention discloses an advertising fusion recommendation algorithm based on operations research optimization, comprising: when a user enters a product recommendation page, obtaining a non-advertising product recommendation queue and an advertising product recommendation queue; using a mixed ranking score calculation module to calculate the mixed ranking score of each advertising product and each non-advertising product respectively; traversing the non-advertising product recommendation queue and the advertising product recommendation queue, using a fusion module to determine the products to be displayed in each product exposure slot on the recommendation page based on the mixed ranking scores of each advertising product and each non-advertising product, thus obtaining the product recommendation result for this user. This invention can, in real time, calculate the most suitable insertion position of advertising and non-advertising products for each user's display request based on advertising supply and the degree of matching with user interests, dynamically adjusting the exposure position of advertising products for each user, effectively improving user experience and advertising revenue.

Description

Technical Field

[0001] This invention relates to the field of advertising recommendation technology, and specifically to an advertising fusion recommendation algorithm based on operations research optimization. Background Technology

[0002] The product recommendation feed in the Weidian APP contains both non-advertised and advertised products. Advertised products refer to those for which merchants have paid for advertising; non-advertised products refer to those for which merchants have not paid for advertising. Currently, advertised products are recommended using fixed slots, such as displaying them at positions 1, 4, and 9 in the recommendation feed.

[0003] The above product recommendation methods have the following shortcomings: (1) Reduced user experience: Abruptly inserting product recommendations will cause a decline in user experience. (2) Unfavorable to platform advertising revenue: When user preferences and product recommendations highly overlap, displaying product recommendations only in fixed slots is not conducive to platform advertising revenue. (3) Lack of personalization: The number of ads and slots displayed to all users are the same. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an advertising fusion recommendation algorithm based on operations research optimization, which can effectively solve the above problems.

[0005] The technical solution adopted in this invention is as follows:

[0006] This invention provides an advertising fusion recommendation algorithm based on operations research optimization, comprising the following steps:

[0007] Step S1: When a user enters a product recommendation page, a request is sent to display the product. When a product display request is made, the non-advertising product recommendation service and the advertising product recommendation service are invoked; the non-advertising product recommendation service returns... The most relevant non-advertised products to the user are ranked from highest to lowest relevance to form a non-advertised product recommendation queue; the advertised product recommendation service returns... The advertising products with the highest revenue on each platform are arranged in descending order of revenue to form a recommendation queue for advertising products.

[0008] Step S2: Using the mixed ranking score calculation module, the mixed ranking score of each recommended advertised product and the mixed ranking score of each non-advertised product are calculated respectively.

[0009] Step S3: Traverse the non-advertising product recommendation queue and the advertising product recommendation queue, and use the fusion module to determine the products to be displayed in each product exposure slot on the recommendation page based on the mixed sorting scores of each advertising product and the mixed sorting scores of each non-advertising product, so as to obtain the product recommendation results for this user.

[0010] Furthermore, step S2 specifically involves:

[0011] Step S21: Using formula (1), obtain the mixed ranking score for each recommended product. The recommended products include both advertised and non-advertised products.

[0012]

[0013] in:

[0014] The cost-per-click bid is calculated per recommended product, in cents. If it's an advertised product, the cost-per-click bid for advertised products will apply; otherwise, it will be calculated based on the cost-per-click bid for non-advertised products. =0;

[0015] Estimated online click-through rate for recommended products;

[0016] As an influencing factor, it is a control parameter dynamically calculated from the supply of advertised goods and the degree of matching with user interests;

[0017]

[0018] in:

[0019] This is the click-through rate ratio for natural ads; it is calculated using the formula: non-ad click-through rate / ad click-through rate / 100, based on the click-through rate of recommended products within a preset time period in the app.

[0020] e is the base of the natural logarithm;

[0021] The overlap score between advertised product keywords and non-advertised product keywords in the non-advertised product recommendation queue and the advertised product recommendation queue is calculated as: number of times advertised product keywords overlap with non-advertised product keywords / number of products displayed per request;

[0022] This refers to the offline calculation of all user requests within the app over a preset time period. The mean.

[0023] Furthermore, step S3 specifically includes:

[0024] Step S31: Take the first advertised product from the advertised product recommendation queue and the first non-advertised product from the non-advertised product recommendation queue;

[0025] Step S32: Compare the mixed sorting scores of advertised and non-advertised products; if the mixed sorting score of non-advertised products is greater than that of advertised products, further determine whether the difference between the position of non-advertised products and the previous position is greater than the set interval parameter theta. If so, insert the non-advertised products into the recommendation result queue; otherwise, insert the advertised products into the recommendation result queue.

[0026] If the ranking score of non-advertised products is not greater than the ranking score of advertised products, then the advertised products will be inserted into the recommendation result queue.

[0027] Step S33: Update the advertising product recommendation queue and the non-advertising product recommendation queue, and remove the corresponding products that were inserted into the recommendation result queue from the advertising product recommendation queue or the non-advertising product recommendation queue;

[0028] Step S34: Repeat steps S31 to S33 until the number of products in the recommendation result queue reaches the specified number of products to be displayed.

[0029] The advertising fusion recommendation algorithm based on operations research optimization provided by this invention has the following advantages:

[0030] This invention provides an advertising fusion recommendation algorithm based on operations research optimization. It can calculate the most suitable insertion position of advertising products and non-advertising products for each user's display request in real time based on the advertising supply and the degree of matching with user interests. It can dynamically adjust the exposure position of advertising products for each user, effectively improving user experience and advertising revenue. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 A flowchart of an advertising fusion recommendation algorithm based on operations research optimization provided by the present invention;

[0033] Figure 2 A flowchart of the fusion module provided by the present invention. Detailed Implementation

[0034] To make the technical problems solved, the technical solutions, and the beneficial effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the invention.

[0035] This invention provides an advertising fusion recommendation algorithm based on operations research optimization. It can calculate the most suitable insertion position of advertising products and non-advertising products for each user's display request in real time based on the advertising supply and the degree of matching with user interests. It can dynamically adjust the exposure position of advertising products for each user, effectively improving user experience and advertising revenue.

[0036] This invention provides an advertising fusion recommendation algorithm based on operations research optimization, such as... Figure 1 As shown, it includes:

[0037] Step S1: When a user enters a product recommendation page, a request is sent to display the product. When a product display request is made, the non-advertising product recommendation service and the advertising product recommendation service are invoked; the non-advertising product recommendation service returns... The most relevant non-advertised products to the user are ranked from highest to lowest relevance to form a non-advertised product recommendation queue; the advertised product recommendation service returns... The advertising products with the highest revenue on each platform are arranged in descending order of revenue to form a recommendation queue for advertising products.

[0038] Step S2: Using the mixed ranking score calculation module, the mixed ranking score of each recommended advertised product and the mixed ranking score of each non-advertised product are calculated respectively.

[0039] Step S21: Using formula (1), obtain the mixed ranking score for each recommended product. The recommended products include both advertised and non-advertised products.

[0040]

[0041] in:

[0042] The cost-per-click bid is calculated per recommended product, in cents. If it's an advertised product, the cost-per-click bid for advertised products will apply; otherwise, it will be calculated based on the cost-per-click bid for non-advertised products. =0;

[0043] Estimated online click-through rate for recommended products;

[0044] As an influencing factor, it is a control parameter dynamically calculated from the supply of advertised goods and the degree of matching with user interests;

[0045]

[0046] in:

[0047] This is the click-through rate ratio for natural ads; it is calculated using the formula: non-ad click-through rate / ad click-through rate / 100, based on the click-through rate of recommended products within a preset time period in the app.

[0048] e is the base of the natural logarithm;

[0049] The overlap score between advertised product keywords and non-advertised product keywords in the non-advertised product recommendation queue and the advertised product recommendation queue is calculated as: number of times advertised product keywords overlap with non-advertised product keywords / number of products displayed per request;

[0050] This refers to the offline calculation of all user requests within the app over a preset time period. The mean.

[0051] Step S3: Traverse the non-advertising product recommendation queue and the advertising product recommendation queue, and use the fusion module to determine the products to be displayed in each product exposure slot on the recommendation page based on the mixed sorting scores of each advertising product and the mixed sorting scores of each non-advertising product, so as to obtain the product recommendation results for this user.

[0052] Step S31: Take the first advertised product from the advertised product recommendation queue and the first non-advertised product from the non-advertised product recommendation queue;

[0053] Step S32: Compare the mixed sorting scores of advertised and non-advertised products; if the mixed sorting score of non-advertised products is greater than that of advertised products, further determine whether the difference between the position of non-advertised products and the previous position is greater than the set interval parameter theta. If so, insert the non-advertised products into the recommendation result queue; otherwise, insert the advertised products into the recommendation result queue.

[0054] If the ranking score of non-advertised products is not greater than the ranking score of advertised products, then the advertised products will be inserted into the recommendation result queue.

[0055] Step S33: Update the advertising product recommendation queue and the non-advertising product recommendation queue, and remove the corresponding products that were inserted into the recommendation result queue from the advertising product recommendation queue or the non-advertising product recommendation queue;

[0056] Step S34: Repeat steps S31 to S33 until the number of products in the recommendation result queue reaches the specified number of products to be displayed.

[0057] The following is an example:

[0058] 1. When a user enters a recommendation page, a response is required for each request. When a product recommendation request is triggered, both the non-advertised product recommendation service and the advertised product recommendation service will be invoked simultaneously. The non-advertised product recommendation service will return... The ad recommendation service will return the most relevant non-advertised product recommendations to the user. The most profitable products on the platform are those recommended by advertisers.

[0059] II. Reasoning of the mixed-sorting formula in the mixed-sorting calculation module:

[0060] The index of the exposure slot is The requested index is The expected engagement utility (which can be understood as the number of clicks on an item) for non-advertised recommended products is: The expected engagement utility of advertised products is The expected benefit utility of advertised recommended products is Expected benefit utility of non-advertised recommended products It is 0.

[0061] For each request They all Each advertisement recommends candidate products and One non-advertised recommended product. For each exposure slot... , This refers to whether or not to place advertisements. The problem is defined as maximizing business metrics while ensuring user-side metrics are maintained.

[0062] Objective function:

[0063] Where: w: quadratic regularization penalty coefficient, which controls the weight of the quadratic penalty term in the objective function, forcing the entire optimization problem to become a strongly convex problem, ensuring that the algorithm can calculate a unique and stable optimal solution. X: full-exposure slot ad placement decision vector, which is the overall decision set / vector composed of the decision variables xᵢ of all exposure slots i, representing the final mixed layout of the entire recommendation page, that is: whether to place ad products or non-ad products in each exposure slot.

[0064] Constraints:

[0065] When all participants are greater than or equal to a certain constant Based on this premise, maximizing the overall benefit, the quadratic term in the objective function is to ensure the strong convexity of the entire problem. Using Lagrange duality, the optimal solution is...

[0066]

[0067] Where: parameters The optimal Lagrangian dual variable, i.e., whether an ad or non-ad is inserted at a certain position, is determined by the advertising revenue of the product. and user effect Composition, advertising revenue General use Measure user benefits Click-through rate is generally used. Therefore, the sorting formula can be defined as follows:

[0068]

[0069] The meanings of each parameter are as follows:

[0070] (1) Bidding price per click for a single ad, in cents, not for ad clicks. =0;

[0071] (2) Estimate click-through rate for products online;

[0072] (3) Impact factor;

[0073] The control parameters are dynamically calculated based on the advertising supply and its match with user interests. The formula is designed as follows:

[0074]

[0075] in:

[0076] (1) This is calculated based on the total click-through rate (non-ad recommendation result click-through rate / ad recommendation result click-through rate / 100) within the app.

[0077] (2) e is the base of the natural logarithm, approximately equal to 2.71828;

[0078] (3) The overlap score is calculated for the product keywords corresponding to the two recommended candidate products: those not advertised and those advertised. The formula is: Number of times the product keyword of the advertised product overlaps with the product keyword of the non-advertised product / Number of products requested per request. In the following example, the overlap score = 5 (Number of times the advertised product overlaps with the non-advertised product keyword) / 10 (Number of products requested) = 0.5

[0079]

[0080] (4) Calculated offline for all requests from all users within the app The mean.

[0081] Therefore, the overall sorting formula is:

[0082]

[0083] For each candidate product that is advertised or not, calculate a mixed ranking score according to the formula above.

[0084] III. Fusion Module: such as Figure 2 As shown, the process iterates through the ad recommendation queue and the non-ad recommendation queue, taking the first item in each queue for comparison. Larger products are added to the final results queue. Simultaneously, to dynamically adjust the ad exposure on each recommendation page, a parameter theta is set for the interval between non-ad appearances (the default interval is 1, meaning an ad product must be inserted between two regular products).

[0085] The steps are as follows:

[0086] (1) Each time, take the first item from the ad recommendation queue and the non-ad recommendation queue.

[0087] (2) Compare the mixed scores of non-advertised products and advertised products. If the mixed score of non-advertised products is greater than that of advertised products, and the difference between the position of non-advertised products and the previous position is greater than the set interval parameter theta, then insert the non-advertised products into the result queue; otherwise, insert the advertised products into the result queue.

[0088] (3) Remove the products that were inserted into the result queue in step 2 from the advertising or non-advertising recommendation queue.

[0089] (4) Repeat steps 1 to 3 until the number of items in the result queue reaches the specified number of returned results.

[0090] The advertising fusion recommendation algorithm based on operations research optimization provided by this invention has the following characteristics:

[0091] 1. Design a novel formula for calculating traffic fusion and mixing.

[0092] 2. A scheme for dynamically adjusting calculation parameters in real time based on the supply of advertisements and their matching degree with user interests is proposed.

[0093] 3. Design a mixed-row calculation parameter flow that can dynamically calculate and calibrate based on the feedback results from the main board without the need for manual verification.

[0094] 4. A method is proposed that uses queue comparison and dynamically inserts the most suitable advertising product position based on the real-time calculated mixed score.

[0095] The advertising fusion recommendation algorithm based on operations research optimization provided by this invention has the following advantages:

[0096] 1. This invention can calculate the most suitable insertion position of the advertised product in real time for each user's display request based on the advertising supply and the degree of matching with user interests, and dynamically adjust the ad exposure position for each user, effectively improving user experience and advertising revenue.

[0097] 2. Based on the feedback from the overall system, parameters can be dynamically calculated and calibrated without manual verification, enabling personalized traffic integration and providing a clear and direct recommendation path with strong interpretability.

[0098] When this traffic fusion method was applied to the product flow recommendation in Weidian, the click-through rate increased by about 4% compared to the benchmark, and advertising revenue increased by 5%.

[0099] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An operational optimization-based advertisement fusion recommendation algorithm, characterized in that, Includes the following steps: Step S1: When a user enters a product recommendation page, a request is sent to display the product. When a product display request is made, the non-advertising product recommendation service and the advertising product recommendation service are invoked; the non-advertising product recommendation service returns... The non-advertised products that are most relevant to the user are arranged in descending order of relevance to form a non-advertised product recommendation queue. The advertising commodity recommendation service returns The advertising commodities are ranked in descending order of the maximum profit of the platform, forming an advertising commodity recommendation queue. Step S2: Using the mixed ranking score calculation module, the mixed ranking score of each recommended advertised product and the mixed ranking score of each non-advertised product are calculated respectively. Step S3: Traverse the non-advertising product recommendation queue and the advertising product recommendation queue, and use the fusion module to determine the products to be displayed in each product exposure slot on the recommendation page based on the mixed sorting scores of each advertising product and the mixed sorting scores of each non-advertising product, so as to obtain the product recommendation results for this user. 2.The algorithm according to claim 1, wherein, Step S2 is as follows: Step S21, using formula (1), obtaining the mixed arrangement ranking score of each recommended commodity ; wherein the recommended commodities include recommended advertising commodities and non-advertising commodities; ； in: The fee-per-click bid for a single recommended commodity, in units of cents, is determined according to the fee-per-click bid of the advertised commodity if the recommended commodity is an advertised commodity, or the fee-per-click bid of the non-advertised commodity if the recommended commodity is a non-advertised commodity. is 0. to recommend an online estimated click rate of a commodity; As an influencing factor, the regulation parameter is dynamically calculated according to the supply of the advertised commodity and the matching degree with the user interest. ； in: is the natural advertisement click rate; the non-advertisement recommendation result click rate and the advertisement recommendation result click rate in the preset period in the statistical market APP are calculated by the formula non-advertisement recommendation result click rate / advertisement recommendation result click rate / 100; e is the base of the natural logarithm; For the non-advertising commodity recommendation queue and the advertising commodity recommendation queue, the coincidence degree of the advertising commodity product word in the non-advertising commodity product word is calculated according to the formula: the number of coincidences of the advertising commodity product word in the non-advertising commodity product word / the number of displayed commodities per request. For a preset time period, the average of all user requests in the APP of the overall market is calculated offline The average of all user requests in the APP of the overall market is calculated offline 3. The advertising fusion recommendation algorithm based on operations research optimization according to claim 1, characterized in that, Step S3 is as follows: Step S31: Take the first advertised product from the advertised product recommendation queue and the first non-advertised product from the non-advertised product recommendation queue; Step S32: Compare the mixed sorting scores of advertised and non-advertised products; if the mixed sorting score of non-advertised products is greater than that of advertised products, further determine whether the difference between the position of non-advertised products and the previous position is greater than the set interval parameter theta. If so, insert the non-advertised products into the recommendation result queue; otherwise, insert the advertised products into the recommendation result queue. If the ranking score of non-advertised products is not greater than the ranking score of advertised products, then the advertised products will be inserted into the recommendation result queue. Step S33: Update the advertising product recommendation queue and the non-advertising product recommendation queue, and remove the corresponding products that were inserted into the recommendation result queue from the advertising product recommendation queue or the non-advertising product recommendation queue; Step S34: Repeat steps S31 to S33 until the number of products in the recommendation result queue reaches the specified number of products to be displayed.

Images