CN-122020078-A - Search relevance evaluation method, device, equipment and storage medium

CN122020078ACN 122020078 ACN122020078 ACN 122020078ACN-122020078-A

Abstract

The technical scheme provided by the embodiment of the specification is used for acquiring search content items, object information of a search object corresponding to the search content items and sub-object information of sub-objects associated with the search object, so as to acquire materials for evaluating the search relevance. Based on the search content item, the object information, and the sub-object information, target hint content is generated that is used to indicate an evaluation of search relevance between the search content item and the search object. The target prompt content is input into a relevance evaluation model, and the relevance evaluation model is used for processing the target prompt content, so that the evaluation of search relevance between the search content item and the search result is realized, and the efficiency of search relevance evaluation is greatly improved. Meanwhile, the dependence on manpower in the search relevance evaluation process is reduced, and the cost is reduced.

Inventors

LI HANG
WANG XIAOWEI
SHI FENG
ZHANG DAN
SUN HUIYE
FANG ZHIJIA

Assignees

拉扎斯网络科技（上海）有限公司

Dates

Publication Date: 20260512
Application Date: 20241111

Claims (10)

1. A method of evaluating search relevance, the method comprising: acquiring a search content item, object information of a search object corresponding to the search content item and sub-object information of a sub-object associated with the search object; Generating target prompt content based on the search content item, the object information and sub-object information of sub-objects associated with the search object, wherein the target prompt content is used for indicating evaluation of search relevance between the search content item and the search object; And inputting the target prompt content into a relevance evaluation model, and processing the target prompt content through the relevance evaluation model to obtain a relevance score between the search content item and the search object, wherein the relevance score is positively correlated with the search relevance between the search content item and the search object.
2. The method of claim 1, the generating target hint content based on the search content item, the object information, and sub-object information of a sub-object associated with the search object, comprising: And fusing the search content item, the object information and the sub-object information of the sub-object associated with the search object with a prompt content template to obtain the target prompt content, wherein the prompt content template comprises a scoring rule when performing relevance evaluation.
3. The method of claim 1, wherein the fusing the search content item, the object information, and the sub-object information of the sub-object associated with the search object with a hint content template to obtain the target hint content includes: under the condition that the search content item, the object information and the sub-object information are all texts, filling the search content item, the object information and the sub-object information into the prompt content template to obtain the target prompt content; And under the condition that non-text exists in the search content item, the object information and the sub-object information, extracting features of the search content item, the object information, the sub-object information and the prompt content template to obtain content item features of the search content item, object information features of the object information, sub-object information features of the sub-object information and content template features of the prompt content template, and carrying out feature fusion on the content item features, the object information features, the sub-object information features and the content template features to obtain the target prompt content.
4. The method of claim 1, the inputting the target prompt into a relevance evaluation model, processing the target prompt through the relevance evaluation model to obtain a relevance score between the search content item and the search object, comprising: Inputting the target prompt content into a relevance evaluation model, and coding the target prompt content based on an attention mechanism through the relevance evaluation model to obtain the prompt content semantic features of the target prompt content; And performing multi-round iterative decoding on the prompt content semantic features of the target prompt content based on an attention mechanism through the relevance evaluation model to obtain relevance scores between the search content items and the search objects.
5. The method according to claim 1, wherein the target prompt includes a scoring rule when performing relevance evaluation, the inputting the target prompt into a relevance evaluation model, and processing the target prompt through the relevance evaluation model, to obtain a relevance score between the search content item and the search object includes: Inputting the target prompt content into a relevance evaluation model, and coding search content items in the target prompt content based on an attention mechanism through the relevance evaluation model to obtain search content item semantic features of the search content items; Encoding object information in the target prompt content and sub-object information of the sub-object associated with the search object based on an attention mechanism through the relevance evaluation model to obtain search object semantic features of the search object; And determining a relevance score between the search content item and the search object based on the scoring rule, the search content item semantic feature of the search content item and the search object semantic feature of the search object through the relevance evaluation model.
6. The method of claim 5, the scoring rules comprising a plurality of candidate relevance instances and candidate relevance scores corresponding to each of the candidate relevance instances, the determining a relevance score between the search content item and the search object based on the scoring rules, search content item semantic features of the search content item, and search object semantic features of the search object comprising: Determining a target relevance situation according to which the search content item and the search object meet in the candidate relevance situations based on feature similarity between the search content item semantic features of the search content item and the search object semantic features of the search object; And determining the candidate relevance scores corresponding to the target relevance conditions as the relevance scores between the search content items and the search objects.
7. The method of claim 1, the obtaining the search content item, the object information of the search object corresponding to the search content item, and the sub-object information of the sub-object associated with the search object, comprising: And acquiring search content items, object information of search objects corresponding to the search content items and sub-object information of sub-objects associated with the search objects from a search exposure log, wherein the search exposure log comprises a plurality of candidate search content items generated in an AB test process, object information of candidate search objects corresponding to the candidate search content items and sub-object information of sub-objects associated with the candidate search objects.
8. An apparatus for evaluating search relevance, the apparatus comprising: the acquisition module is used for acquiring the search content item, the object information of the search object corresponding to the search content item and the sub-object information of the sub-object associated with the search object; A prompt content generation module for generating target prompt content based on the search content item, the object information and sub-object information of the sub-object associated with the search object, wherein the target prompt content is used for indicating to evaluate search relativity between the search content item and the search object; And the processing module is used for inputting the target prompt content into a relevance evaluation model, and processing the target prompt content through the relevance evaluation model to obtain a relevance score between the search content item and the search object, wherein the relevance score is positively correlated with the search relevance between the search content item and the search object.
9. An electronic device comprising one or more processors and one or more memories, the one or more memories having stored therein at least one computer program loaded and executed by the one or more processors to implement the method of evaluating search relevance of any of claims 1-7.
10. A computer-readable storage medium having stored therein at least one computer program that is loaded and executed by a processor to implement the method of evaluating search relevance of any one of claims 1 to 7.

Description

Search relevance evaluation method, device, equipment and storage medium Technical Field Embodiments of the present disclosure relate to the field of computer technologies, and in particular, to a method, an apparatus, a device, and a storage medium for evaluating search relevance. Background As network technology evolves, more and more users will search for search objects or sub-objects on the network. The search relevance is an important index for measuring the quality of the search result, and for users, the search is a process of finding the wanted result with clear intention, if the obtained search result is an irrelevant search object or sub-object, the user experience can be greatly hurt, the final order conversion is influenced, and even the user loss is directly caused. Therefore, evaluating search relevance is an important part in optimizing search performance. In the related art, the search relevance is often marked manually, that is, the search content items and the search results are sent to a marking person, and the marking person marks the search relevance between each group of the search content items and the search results manually. However, the manual labeling method is low in efficiency and high in cost. Disclosure of Invention The embodiment of the specification provides a method, a device, equipment and a storage medium for evaluating search relevance, which can improve the efficiency of evaluating search relevance and reduce the cost, and the technical scheme is as follows: in one aspect, a method for evaluating search relevance is provided, the method comprising: acquiring a search content item, object information of a search object corresponding to the search content item and sub-object information of a sub-object associated with the search object; Generating target prompt content based on the search content item, the object information and sub-object information of sub-objects associated with the search object, wherein the target prompt content is used for indicating evaluation of search relevance between the search content item and the search object; And inputting the target prompt content into a relevance evaluation model, and processing the target prompt content through the relevance evaluation model to obtain a relevance score between the search content item and the search object, wherein the relevance score is positively correlated with the search relevance between the search content item and the search object. In one aspect, there is provided an apparatus for evaluating search relevance, the apparatus including: the acquisition module is used for acquiring the search content item, the object information of the search object corresponding to the search content item and the sub-object information of the sub-object associated with the search object; A prompt content generation module for generating target prompt content based on the search content item, the object information and sub-object information of the sub-object associated with the search object, wherein the target prompt content is used for indicating to evaluate search relativity between the search content item and the search object; And the processing module is used for inputting the target prompt content into a relevance evaluation model, and processing the target prompt content through the relevance evaluation model to obtain a relevance score between the search content item and the search object, wherein the relevance score is positively correlated with the search relevance between the search content item and the search object. In a possible implementation manner, the prompt content generation module is configured to fuse the search content item, the object information, and sub-object information of a sub-object associated with the search object with a prompt content template, to obtain the target prompt content, where the prompt content template includes a scoring rule when performing relevance evaluation. In one possible implementation manner, the prompt content generation module is configured to populate the search content item, the object information, and the sub-object information into the prompt content template to obtain the target prompt content when the search content item, the object information, and the sub-object information are text, and perform feature extraction on the search content item, the object information, the sub-object information, and the prompt content template to obtain a content item feature of the search content item, an object information feature of the object information, a sub-object information feature of the sub-object information, and a content template feature of the prompt content template when the search content item, the object information, and the sub-object information feature of the sub-object information exist in the non-text, and perform feature fusion on the content item feature, the object information feature, the sub-object information feature, and the content template