CN-121998731-A - Information pushing method, model training method, device, equipment and storage medium

Abstract

The embodiment of the application provides an information pushing method, a model training method, a device, equipment and a storage medium. The method comprises the steps of obtaining commodity multi-source data of commodities, extracting characteristics of the commodity multi-source data of the commodities to obtain total characteristic vectors of the commodities, carrying out sharing coding processing on the total characteristic vectors of the commodities based on a sharing coder of a multi-task model to obtain sharing representation vectors of the commodities, wherein the sharing representation vectors represent context semantic information of the commodities, processing the sharing representation vectors of the commodities based on a multi-task head of the multi-task model to obtain forecast information of the commodities in a future time period, wherein the forecast information comprises at least one of pricing advice, comprehensive success probability and demand change trend, and carrying out information pushing processing on the forecast information of the commodities. The method can improve the transaction efficiency and the user satisfaction of the second-hand commodity transaction platform.

Inventors

• WU YUHAN

Assignees

• 北京侠客汇信息技术有限责任公司

Dates

Publication Date

20260508

Application Date

20251223

Claims (19)

1. 1. An information pushing method is characterized by comprising the following steps: extracting characteristics of the commodity multi-source data of the commodity to obtain a full characteristic vector of the commodity; The shared encoder based on the multi-task model carries out shared encoding processing on the full feature vector of the commodity to obtain a shared representation vector of the commodity, wherein the shared representation vector represents the context semantic information of the commodity; Processing the shared representation vector of the commodity based on the multi-task head of the multi-task model to obtain the prediction information of the commodity in a future time period, wherein the prediction information comprises at least one of pricing advice, comprehensive success probability and demand change trend; And carrying out information pushing processing on the predicted information of the commodity.
2. 2. The method of claim 1, wherein the feature extracting the multi-source data of the commodity to obtain a full feature vector of the commodity comprises: Extracting features of internal data in the commodity multi-source data to obtain an internal feature vector of the commodity, wherein the internal data comprises commodity static properties, commodity dynamic data, transaction data and user data; extracting features of external data in the commodity multi-source data to obtain external feature vectors of the commodity, wherein the external data comprises platform data, public opinion data and market data; and splicing the internal feature vector and the external feature vector of the commodity to obtain the full feature vector of the commodity.
3. 3. The method of claim 2, wherein the internal feature vectors include structured feature vectors and unstructured feature vectors, wherein the structured feature vectors include numerical features, category features, and time-series features; and/or the external feature vector comprises a key ratio feature, a public opinion feature and a quadruple.
4. 4. The method according to claim 1, wherein the processing the shared representation vector of the commodity based on the multitasking head of the multitasking model to obtain the predicted information of the commodity in the future time period includes: based on the pricing regression head of the multitask head, processing the shared representation vector to obtain pricing advice in the forecast information of the commodity; And/or processing the shared representation vector and the user intention vector in the full feature vector based on the supply and demand matching head of the multi-task head to obtain the comprehensive success probability in the prediction information of the commodity; and/or processing the shared representation vector based on the trend prediction head of the multi-task head to obtain the demand change trend in the predicted information of the commodity.
5. 5. The method of claim 4, wherein the processing the user intention vector of the shared representation vector and the full feature vector based on the supply-demand matching head of the multi-tasking head to obtain the integrated success probability in the predicted information of the commodity comprises: determining a similarity between the shared representation vector and a user intent vector in the full feature vector based on the supply-demand matching head; Calculating pricing suggestions and current prices in the prediction information of the commodity to obtain a rationality score; And based on the fully-connected network of the supply and demand matching head, carrying out fusion processing on the similarity, the rationality score and the seller qualification rate in the full feature vector to obtain the comprehensive success probability in the predicted information of the commodity.
6. 6. The method of claim 4, wherein the processing the shared representation vector based on the trend prediction head of the multitasking head to obtain the trend of demand change in the predicted information of the commodity comprises: the method comprises the steps of carrying out coding processing on a historical sequence of the commodity to obtain a time sequence context, wherein the historical sequence comprises historical prices and volume of transactions of the commodity in a historical time period; and based on the trend predicting head, carrying out fusion prediction on the time sequence context and the shared representation vector to obtain the demand change trend in the predicted information of the commodity.
7. 7. The method according to any one of claims 1 to 6, wherein the information pushing processing of the predicted information of the commodity includes: If the change value corresponding to the change trend of the demand represented by the predicted information is larger than a preset threshold value, generating a push message of the commodity according to the predicted information of the commodity, wherein the push message comprises the change trend of the demand and a jump link; and carrying out asynchronous pushing processing on the push message.
8. 8. The model training method applied to information push is characterized by comprising the following steps of: The method comprises the steps of obtaining commodity multi-source data of at least one commodity to be trained, and extracting characteristics of the commodity multi-source data of the commodity to be trained to obtain a full-quantity characteristic vector of the commodity to be trained; Carrying out shared coding processing on the full feature vector of the commodity to be trained based on a shared coder of an initial model to obtain a shared representation vector of the commodity to be trained, wherein the shared representation vector represents context semantic information of the commodity to be trained; Processing the shared representation vector of the commodity to be trained based on the multi-task head of the initial model to obtain the prediction information of the commodity to be trained in a preset time period, wherein the prediction information comprises at least one of pricing advice, comprehensive success probability and demand change trend; Training the initial model according to the prediction information of the commodity to be trained to obtain a multi-task model, wherein the multi-task model is used for processing commodity multi-source data of the commodity according to any one of claims 1-7 to obtain the prediction information of the commodity.
9. 9. The method of claim 8, wherein the feature extraction of the multi-source data of the to-be-trained commodity to obtain a full feature vector of the to-be-trained commodity comprises: extracting features of internal data in the commodity multi-source data to obtain an internal feature vector of the commodity to be trained, wherein the internal data comprises commodity static properties, commodity dynamic data, transaction data and user data; extracting features of external data in the commodity multi-source data to obtain external feature vectors of the commodities to be trained, wherein the external data comprises platform data, public opinion data and market data; And splicing the internal feature vector and the external feature vector of the commodity to be trained to obtain the full feature vector of the commodity to be trained.
10. 10. The method of claim 9, wherein the internal feature vectors include structured feature vectors and unstructured feature vectors, wherein the structured feature vectors include numerical features, category features, and time-series features; and/or the external feature vector comprises a key ratio feature, a public opinion feature and a quadruple.
11. 11. The method of claim 8, wherein the processing the shared representation vector of the to-be-trained commodity based on the multitasking header of the initial model to obtain the predicted information of the to-be-trained commodity within a preset time period includes: Based on the pricing regression head of the multitask head, processing the shared representation vector to obtain pricing advice in the forecast information of the commodity to be trained; And/or processing the shared representation vector and the user intention vector in the full feature vector based on the supply and demand matching head of the multi-task head to obtain the comprehensive success probability in the prediction information of the commodity to be trained; and/or processing the shared representation vector based on the trend prediction head of the multitask head to obtain the demand change trend in the predicted information of the commodity to be trained.
12. 12. The method of claim 11, wherein the processing the user intention vector of the shared representation vector and the full feature vector based on the supply-demand matching head of the multi-task head to obtain the comprehensive success probability in the predicted information of the commodity to be trained comprises: determining a similarity between the shared representation vector and a user intent vector in the full feature vector based on the supply-demand matching head; calculating pricing suggestions and current prices in the forecast information of the goods to be trained to obtain a rationality score; And based on the fully-connected network of the supply and demand matching head, carrying out fusion processing on the similarity, the rationality score and the seller qualification rate in the full feature vector to obtain the comprehensive success probability in the forecast information of the commodity to be trained.
13. 13. The method of claim 11, wherein the processing the shared representation vector based on the trend prediction head of the multitask header to obtain the demand change trend in the predicted information of the commodity to be trained comprises: Coding the historical sequence of the commodity to be trained to obtain a time sequence context, wherein the historical sequence comprises the historical price and the volume of the commodity to be trained in a historical time period; And based on the trend predicting head, carrying out fusion prediction on the time sequence context and the shared representation vector to obtain the demand change trend in the predicted information of the commodity to be trained.
14. 14. The method according to any one of claims 8-13, wherein training the initial model according to the prediction information of the commodity to be trained to obtain a multi-task model includes: Determining a pricing loss function according to the actual pricing result of the commodity to be trained and the pricing advice in the forecast information; Determining a matching loss function according to the actual comprehensive success probability of the commodity to be trained and the comprehensive success probability in the prediction information; Determining a trend loss function according to the actual demand change trend of the commodity to be trained and the demand change trend in the forecast information; determining a total loss function from the pricing loss function, the matching loss function, and the trend loss function; Training the initial model according to the total loss function to obtain the multi-task model.
15. 15. An information pushing apparatus, characterized by comprising: the extraction module is used for acquiring commodity multi-source data of the commodity, and extracting characteristics of the commodity multi-source data of the commodity to obtain a full characteristic vector of the commodity; the coding module is used for carrying out shared coding processing on the total feature vector of the commodity based on a shared coder of the multi-task model to obtain a shared representation vector of the commodity, wherein the shared representation vector represents the context semantic information of the commodity; The prediction module is used for processing the shared representation vector of the commodity based on the multitask head of the multitask model to obtain the prediction information of the commodity in a future time period, wherein the prediction information comprises at least one of pricing advice, comprehensive success probability and demand change trend; and the pushing module is used for carrying out information pushing processing on the predicted information of the commodity.
16. 16. Model training device for information push, characterized by comprising: The device comprises an extraction module, a training module and a training module, wherein the extraction module is used for acquiring commodity multi-source data of at least one commodity to be trained, and extracting characteristics of the commodity multi-source data of the commodity to be trained to obtain a full-quantity characteristic vector of the commodity to be trained; The coding module is used for carrying out shared coding processing on the total feature vector of the commodity to be trained based on the shared coder of the initial model to obtain a shared representation vector of the commodity to be trained, wherein the shared representation vector represents the context semantic information of the commodity to be trained; The prediction module is used for processing the shared representation vector of the commodity to be trained based on the multi-task head of the initial model to obtain the prediction information of the commodity to be trained in a preset time period, wherein the prediction information comprises at least one of pricing advice, comprehensive success probability and demand change trend; the training module is used for training the initial model according to the prediction information of the commodity to be trained to obtain a multi-task model, wherein the multi-task model is used for processing the commodity multi-source data of the commodity according to claim 15 to obtain the prediction information of the commodity.
17. 17. An electronic device is characterized by comprising a memory and a processor; The memory stores computer-executable instructions; the processor executing computer-executable instructions stored in the memory, causing the processor to perform the method of any one of claims 1-14.
18. 18. A computer readable storage medium having stored therein computer executable instructions which when executed by a processor are adapted to carry out the method of any one of claims 1-14.
19. 19. A computer program product, characterized in that the computer program product comprises a computer program which, when executed by a processor, implements the method of any one of claims 1-14.

Description

Information pushing method, model training method, device, equipment and storage medium Technical Field The present application relates to the field of artificial intelligence technologies, and in particular, to an information pushing method, a model training method, a device, equipment, and a storage medium. Background In the current secondhand commodity transaction market, a seller user prices a commodity, a recommendation system recommends the commodity (including the commodity) to the buyer user when the buyer user searches the commodity, and the conventional recommendation system can not be deeply matched by combining key attributes such as commodity pricing rationality and color forming integrity based on a key word matching or basic collaborative filtering algorithm when the buyer is in the process of purchasing, so that a recommendation result is deviated from the actual requirement (such as 'cost performance ratio', 'quick sale') of the user. In the prior art, commodity pricing is carried out based on historical exchange average price, and commodity recommendation is carried out through keyword matching or basic collaborative filtering algorithm. In the method, sellers excessively rely on historical exchange average price in pricing to cause the commodity pricing to be disjointed with the actual market demands, and meanwhile, only rely on keyword matching or collaborative filtering, the matching dimension is single, price rationality is ignored, and recommendation results are disjointed with the user demands. Disclosure of Invention The information pushing method, the model training method, the device, the equipment and the storage medium provided by the application promote the scientificity of dynamic commodity pricing, the accuracy of supply and demand matching and the prospective of market prediction, thereby promoting the transaction efficiency and the user satisfaction of the second-hand commodity transaction platform. In a first aspect, the present application provides an information pushing method, including: extracting characteristics of the commodity multi-source data of the commodity to obtain a full characteristic vector of the commodity; The shared encoder based on the multi-task model carries out shared encoding processing on the full feature vector of the commodity to obtain a shared representation vector of the commodity, wherein the shared representation vector represents the context semantic information of the commodity; Processing the shared representation vector of the commodity based on the multi-task head of the multi-task model to obtain the prediction information of the commodity in a future time period, wherein the prediction information comprises at least one of pricing advice, comprehensive success probability and demand change trend; And carrying out information pushing processing on the predicted information of the commodity. In one possible implementation manner, the feature extraction of the multi-source data of the commodity to obtain a full feature vector of the commodity includes: Extracting features of internal data in the commodity multi-source data to obtain an internal feature vector of the commodity, wherein the internal data comprises commodity static properties, commodity dynamic data, transaction data and user data; extracting features of external data in the commodity multi-source data to obtain external feature vectors of the commodity, wherein the external data comprises platform data, public opinion data and market data; and splicing the internal feature vector and the external feature vector of the commodity to obtain the full feature vector of the commodity. In one possible implementation, the internal feature vector comprises a structured feature vector and an unstructured feature vector, wherein the structured feature vector comprises numerical features, category features and time sequence features; and/or the external feature vector comprises a key ratio feature, a public opinion feature and a quadruple. In one possible implementation manner, the processing, by the multitasking head based on the multitasking model, the shared representation vector of the commodity to obtain the prediction information of the commodity in the future time period includes: based on the pricing regression head of the multitask head, processing the shared representation vector to obtain pricing advice in the forecast information of the commodity; And/or processing the shared representation vector and the user intention vector in the full feature vector based on the supply and demand matching head of the multi-task head to obtain the comprehensive success probability in the prediction information of the commodity; and/or processing the shared representation vector based on the trend prediction head of the multi-task head to obtain the demand change trend in the predicted information of the commodity. In one possible implementation manner, the processing, by the supply and