Search

CN-122019887-A - Method and device for determining distribution user of supply

CN122019887ACN 122019887 ACN122019887 ACN 122019887ACN-122019887-A

Abstract

The embodiment of the specification provides a method and a device for determining distribution users of feeds, wherein the method comprises the steps of determining a plurality of associated historical feeds associated with a first feed from a historical feed set, wherein the historical feed set comprises a plurality of historical feeds and user interaction data corresponding to the historical feeds, the first feed is cold start feed, determining a plurality of candidate users of the first feed based on the user interaction data of the historical feeds, determining audience images corresponding to the first feed through a large language model by utilizing feed information of the first feed, and determining a plurality of target distribution users of the first feed based on the audience images and the candidate users, so that the matched target distribution users are determined for the cold start feed, namely the cold start feed and the target distribution users thereof are efficiently and accurately matched, and the overall conversion efficiency of a digital platform is improved.

Inventors

  • YANG BING

Assignees

  • 支付宝(杭州)数字服务技术有限公司

Dates

Publication Date
20260512
Application Date
20260214

Claims (17)

  1. 1. A method of determining a distribution user of a offer, comprising: determining a number of associated historical feeds associated with a first feed from a set of historical feeds, wherein the set of historical feeds includes a plurality of historical feeds and their respective corresponding user interaction data, the first feed being a cold start feed; determining a number of candidate users of the first offer based on the number of associated historical offered user interaction data; Determining an audience image corresponding to the first supply through a large language model by using the supply information of the first supply; a number of target distribution users of the first offer are determined based on the audience representation and the number of candidate users.
  2. 2. The method of claim 1, wherein the determining the first plurality of distribution users of the first offer comprises: Screening the candidate users based on the audience portraits to obtain first distribution users of the first supply; the number of target distribution users is determined based on the number of first distribution users.
  3. 3. The method of claim 2, wherein the determining the number of target distribution users comprises: determining the current user characterization of each first distribution user; Determining a first supply characterization of the first supply; Determining a plurality of target distribution users of the first offer based on each current user representation of each first distribution user and the first offer representations using a reinforcement learning model, wherein the reinforcement learning model is trained based on a reinforcement learning algorithm.
  4. 4. The method of claim 2, wherein the first plurality of distribution users that obtain the first offer comprise: converting the audience image into a screening logic rule; And screening the plurality of first distribution users from the plurality of candidate users based on the screening logic rules.
  5. 5. The method of claim 1, wherein the determining a number of association history feeds associated with the first feed comprises: Determining a first supply characterization of the first supply based on the supply information of the first supply; Based on the first supply characterization, a number of associated historical supplies associated with the first supply are determined from the set of historical supplies.
  6. 6. The method of claim 5, wherein the determining the first supply characterization thereof comprises: extracting text content, visual content and structured content from the feed information; Respectively extracting characteristics of the text content, the visual content and the structured content to obtain respective characteristic vectors; And fusing the feature vectors to obtain the first supply representation.
  7. 7. The method of claim 5, wherein the historical supply set further comprises a plurality of second supply characterizations corresponding to a plurality of historical supplies; the determining a number of associated history feeds associated with the first feed includes: determining semantic similarity between the first offer representation and each second offer representation; and taking the historical supplies corresponding to the K second supply characterizations with the highest semantic similarity as the plurality of associated historical supplies.
  8. 8. The method of claim 1, wherein the determining a number of candidate users for the first offer comprises: For a first history feed of any of the plurality of associated history feeds, determining an effective distribution channel for the first history feed based on a plurality of distribution channels to which user interaction data for the first history feed relates; The number of candidate users is determined based on the users involved in the effective distribution channels of the respective associated history feeds.
  9. 9. The method of claim 8, wherein the determining the effective distribution channel of the first historical offer comprises: Based on the user interaction data provided by the first history, counting the conversion performance of the user interaction data on each distribution channel; The effective distribution channels are screened from all the distribution channels involved based on their conversion performance on each distribution channel.
  10. 10. The method of claim 3, wherein the reinforcement learning model comprises a policy network; said determining a number of target distribution users of said first offer, comprising: Based on the current strategy of the strategy network, screening a plurality of current distribution users of the first supply from the first distribution users according to the current user characterization and the first supply characterization of the first distribution users; Distributing the first offer to each current distribution user, and determining the performance behavior data of each current distribution user on the first offer; determining a current prize value based on performance behavior data of each current distribution user for the first offer; Updating the current policy based at least on the current prize value; and determining the plurality of target distribution users based on the updated policies of the policy network.
  11. 11. The method of claim 10, wherein the screening out a number of currently distributed users of the first offer comprises: Inputting each current user token and the first offer token of each first distribution user into the policy network, so that the policy network processes each current user token and the first offer token based on the current policy thereof to obtain a first probability distribution, wherein the first probability distribution indicates a first probability that each first distribution user is selected as a current distribution user of the first offer; And screening a plurality of current distribution users of the first supply from the plurality of first distribution users based on the first probability distribution.
  12. 12. The method of claim 11, wherein each first distribution user corresponds to a Beta distribution, respectively; The screening out a number of currently distributed users of the first offer includes: Determining a plurality of first users meeting preset conditions from the plurality of first distribution users; Updating each first probability of each first user in the first probability distribution based on Beta distribution corresponding to each first user; And screening the plurality of current distribution users from the plurality of first distribution users based on the updated first probability distribution.
  13. 13. The method of claim 10, wherein the reinforcement learning model comprises a value network; the updating the current policy based at least on the current prize value includes: Determining a first value estimate based on a value network from each current user representation of each first distribution user and the first offer representation; Updating the current policy based on the current prize value and the first price estimate.
  14. 14. The method of claim 13, wherein after the updating the current policy, further comprising: Updating the current user characterization of each current distribution user based on the performance behavior data of each current distribution user on the first supply, thereby obtaining updated current user characterization of each first distribution user, wherein the updated current user characterization of each first distribution user is used for determining a plurality of current distribution users of the first supply in the next iteration process.
  15. 15. The method of claim 1, further comprising: the first offer is distributed to each target distribution user.
  16. 16. An apparatus for determining a served distribution user, comprising: A first determination module configured to determine a number of associated historical feeds associated with a first feed from a set of historical feeds, wherein the set of historical feeds includes a plurality of historical feeds and their respective corresponding user interaction data, the first feed being a cold start feed; a second determination module configured to determine a number of candidate users of the first offer based on the number of associated historical offered user interaction data; a third determining module configured to determine an audience image corresponding to the first offer through a large language model using the offer information of the first offer; A fourth determination module configured to determine a number of target distribution users of the first offer based on the audience representation and the number of candidate users.
  17. 17. A computing device comprising a memory and a processor, wherein the memory has executable code stored therein, which when executed by the processor, implements the method of any of claims 1-15.

Description

Method and device for determining distribution user of supply Technical Field The present disclosure relates to the field of artificial intelligence, and in particular, to a method and apparatus for determining a distribution user of a offer. Background As the internet industry goes from the "incremental users" phase to the "stock users" phase, the newly added users grow significantly slower and the acquisition costs continue to climb, which makes each user touch and content exposure particularly valuable. Meanwhile, various digital feeds (also referred to as content) of goods, videos, images, articles, and advertising creatives on digital platforms have exhibited explosive growth, and a large number of new digital feeds continue to be launched into digital platforms. These new digital feeds that have just been brought up are commonly referred to as cold-start (cold-start) feeds, among others. These cold-start feeds are generally lacking in historical interaction data (e.g., user click, conversion, browsing, etc. interaction data), and under the relevant recommendation or distribution mechanisms, these cold-start feeds are difficult to obtain effective exposure and are easily submerged in massive information, resulting in insufficient release of their potential value. Under the dual pressures of "stock user competition exacerbation" and "supply scale increase", there is a need for an improved method of determining distribution (or recommendation) users of supplies to determine matching distribution users for cold start supplies, improving the overall conversion efficiency of the digital platform. Disclosure of Invention One or more embodiments of the present disclosure provide a method and an apparatus for determining a distribution user of a supply, so as to determine a matched distribution user for a cold start supply, that is, perform efficient and accurate matching between the cold start supply and the distribution user thereof, thereby improving the overall conversion efficiency of a digital platform. According to a first aspect, there is provided a method of determining a dispensing user of a offer, comprising: determining a number of associated historical feeds associated with a first feed from a set of historical feeds, wherein the set of historical feeds includes a plurality of historical feeds and their respective corresponding user interaction data, the first feed being a cold start feed; determining a number of candidate users of the first offer based on the number of associated historical offered user interaction data; Determining an audience image corresponding to the first supply through a large language model by using the supply information of the first supply; a number of target distribution users of the first offer are determined based on the audience representation and the number of candidate users. According to a second aspect, there is provided an apparatus for determining a dispensing user of a offer, comprising: A first determination module configured to determine a number of associated historical feeds associated with a first feed from a set of historical feeds, wherein the set of historical feeds includes a plurality of historical feeds and their respective corresponding user interaction data, the first feed being a cold start feed; a second determination module configured to determine a number of candidate users of the first offer based on the number of associated historical offered user interaction data; a third determining module configured to determine an audience image corresponding to the first offer through a large language model using the offer information of the first offer; A fourth determination module configured to determine a number of target distribution users of the first offer based on the audience representation and the number of candidate users. According to a third aspect, there is provided a computer readable storage medium having stored thereon a computer program which, when executed in a computer, causes the computer to perform the method of the first aspect. According to a fourth aspect, there is provided a computing device comprising a memory and a processor, wherein the memory has executable code stored therein, and wherein the processor, when executing the executable code, implements the method of the first aspect. According to the method and the device, a plurality of relevant historical supplies associated with a first supply are determined from a historical supply set, wherein the historical supply set comprises a plurality of historical supplies and user interaction data corresponding to the historical supplies respectively, the first supply is a cold start supply, a plurality of candidate users of the first supply are determined based on the user interaction data corresponding to the historical supplies, audience images corresponding to the first supply are determined through a large language model by using supply information of the first supply, and a