CN118227751A - Target searching method, target searching device, electronic equipment and readable storage medium - Google Patents

Abstract

The application relates to the field of artificial intelligence, and provides a target searching method, a target searching device, electronic equipment and a readable storage medium, wherein the method comprises the following steps: acquiring a target search text and user context information, and acquiring a target search language chain according to the target search text and the user context information; inputting the target search language chain into a similarity processing model to perform similarity calculation to obtain a first candidate item set output by the similarity processing model; obtaining search prompt information according to the target search language chain and the first candidate item set; inputting the search prompt information into a large language model to obtain a second candidate item set output by the large language model; and obtaining target search results according to the second candidate item set. The application can improve the searching accuracy of the searching system in the small sample field and the cold start field.

Description

Target searching method, target searching device, electronic equipment and readable storage medium

Technical Field

The present application relates to the field of artificial intelligence, and in particular, to a target searching method, apparatus, electronic device, and readable storage medium.

Background

The searching system is used as an important component of the brand resource library and is mainly used for helping users to find required brand assets more quickly, and is mainly characterized by full-text searching function, multidimensional searching, multi-modal searching and the like. The user can quickly and accurately find the required brand assets, and the user experience and the working efficiency are greatly improved.

The processing logic of the search system in the current industry is that query understanding is performed through natural language processing (Natural Language Processing, NLP), thousands of contents are provided to a coarse-rank model through data recall, hundreds of data are provided to a fine-rank model after scoring and rearrangement of the coarse-rank model, tens of groups of data are provided to rearrangement after scoring and processing of the fine-rank model, and finally presented to a user after rearrangement, but the conventional search system usually lacks enough data and user feedback in the fields of cold start and small samples to perform accurate ranking and personalized recommendation, so that the search system cannot provide a good search effect.

Disclosure of Invention

In view of the above, the embodiments of the present application provide a target searching method, apparatus, electronic device, and readable storage medium, so as to solve the problem that in the prior art, high-precision search results cannot be provided in the cold start field and the small sample field.

In a first aspect of an embodiment of the present application, there is provided a target searching method, including:

acquiring a target search text and user context information, and acquiring a target search language chain according to the target search text and the user context information; inputting a target search language chain into a similarity processing model

Performing similarity calculation to obtain a first candidate item set output by a similarity processing model; obtaining search prompt information according to the target search language chain and the first candidate item set; inputting the search prompt information into the large language model to obtain a second candidate item set output by the large language model; and obtaining target search results according to the second candidate item set, wherein the target search results comprise target searches and description information aiming at the target searches.

In a second aspect of an embodiment of the present application, there is provided a target search apparatus including:

The acquisition module is configured to acquire target search text and user context information, and acquire a target search language chain according to the target search text and the user context information; the similarity processing module is configured to input the target search language chain into the similarity processing model to perform similarity calculation, so as to obtain a first candidate item set output by the similarity processing model; the prompting module is configured to obtain search prompting information according to the target search language chain and the first candidate item set; the screening module is configured to input the search prompt information into the large language model to obtain a second candidate item set output by the large language model; and the searching module is configured to obtain target searching results according to the second candidate item set, wherein the target searching results comprise target searches and description information aiming at the target searches.

In a third aspect of the embodiments of the present application, there is provided an electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the steps of the above method when executing the computer program.

In a fourth aspect of the embodiments of the present application, there is provided a computer readable storage medium storing a computer program which, when executed by a processor, implements the steps of the above method.

Compared with the prior art, the embodiment of the application has the beneficial effects that:

the method comprises the steps of obtaining a target search text and user context information, and obtaining a target search language chain according to the target search text and the user context information; inputting the target search language chain into a similarity processing model to perform similarity calculation to obtain a first candidate item set output by the similarity processing model; obtaining search prompt information according to the target search language chain and the first candidate item set; inputting the search prompt information into the large language model to obtain a second candidate item set output by the large language model; and obtaining target search results according to the second candidate item set, wherein the target search results comprise target searches and description information aiming at the target searches. The application improves the searching accuracy of the searching system in the small sample field and the cold start field.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.

Fig. 1 is a schematic flow chart of a target searching method according to an embodiment of the present application;

fig. 2 is a schematic structural diagram of a target searching device according to an embodiment of the present application;

Fig. 3 is a schematic structural diagram of an electronic device according to an embodiment of the present application.

Detailed Description

In the following description, for purposes of explanation and not limitation, specific details are set forth such as the particular system architecture, techniques, etc., in order to provide a thorough understanding of the embodiments of the present application. It will be apparent, however, to one skilled in the art that the present application may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.

The term "including" and variations thereof as used herein are intended to be open-ended, i.e., including, but not limited to. The term "based on" is based at least in part on. The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments. Related definitions of other terms will be given in the description below. It should be noted that the concepts of "first", "second", etc. mentioned in this disclosure are only used to distinguish between different devices, modules or units, and are not intended to limit the order or interdependence of functions performed by these devices, modules or units.

It should be noted that references to "a" and "an" in this disclosure are intended to be illustrative rather than limiting, and those of ordinary skill in the art will appreciate that "one or more" is intended to be understood as "one or more" unless the context clearly indicates otherwise.

The processing logic of the current search system mainly covers links such as Query understanding, recall, coarse ranking, fine ranking and rearrangement.

The Query understanding process comprises preprocessing, word segmentation, rewriting, importance analysis, intention recognition, sensitive recognition and the like, and the Query input by the user is converted into a search tree, so that accurate recall is realized. The recall link is generally formed by fusing multiple recall strategies and is mainly divided into label recall and vector recall, and meanwhile, heat, coverage, correlation and freshness are considered, and thousands of related contents are screened from hundreds of millions of objects. The aim of the coarse discharge is to improve recall accuracy and reduce precision discharge service pressure. Usually, a lightweight machine learning model is adopted to score thousands of contents one by one, and hundreds of objects with the highest scores are selected to enter the next link. Fine-pitch aims at improving flow utilization efficiency and content matching quality, and a large number of characteristic and deep neural network models are usually applied. The rearrangement aims to optimize user experience and content diversity and improve traffic utilization efficiency. Fine tuning is typically performed on a fine-ranking basis and the results presented to the user as a final ranking.

While current search systems have achieved good search results in many ways, most are accurate ranking and personalized recommendations with sufficient data and user feedback, i.e., a large amount of data is needed to support, and if available under a particular domain or topic (small sample domain) or when a new domain or topic appears (cold start domain), the amount of data available is very limited, without the associated training data and user feedback information, the search system may not be able to obtain enough data to train and optimize the model, be hard to understand and satisfy the user's query requirements, resulting in poor search results.

In view of the above, an embodiment of the present application provides a target searching method for solving the above problems.

A target searching method, apparatus, electronic device, and readable storage medium according to embodiments of the present application will be described in detail below with reference to the accompanying drawings.

Fig. 1 is a schematic flow chart of a target searching method according to an embodiment of the present application. As shown in fig. 1, the method includes:

s101, acquiring a target search text and user context information, and acquiring a target search language chain according to the target search text and the user context information;

s102, inputting a target search language chain into a similarity processing model to perform similarity calculation, and obtaining a first candidate item set output by the similarity processing model;

s103, obtaining search prompt information according to the target search language chain and the first candidate item set;

S104, inputting the search prompt information into the large language model to obtain a second candidate item set output by the large language model;

and S105, obtaining target search results according to the second candidate item set, wherein the target search results comprise target searches and description information aiming at the target searches.

Specifically, the above-mentioned target search method is applied to the server and/or the terminal, that is, the above-mentioned target search method may be performed by the server alone or by the terminal alone, the above-mentioned target search method may also be performed by the server and the terminal together, which is not limited in this embodiment, and the present example is described as a target search method applied to the terminal for better description.

The large language model has strong semantic generalization and small sample learning capability, language statistics rules and semantic representation can be obtained through large-scale pre-training, context awareness capability is used for reasoning and prediction, a parameter sharing mechanism transfers previously learned knowledge, and general knowledge is transferred and learned to different tasks, so that the application of the general knowledge to a recommendation system can improve the search accuracy of a search system in the related field. The specific processing logic is as follows:

firstly, when the terminal receives the search requirement of the user, keywords or phrases can be extracted from the search requirement to obtain target search text, and secondly, user context information related to the current situation of the user, such as user portrait, historical interaction information, historical dialogue information and the like, is obtained, so that the search feature capable of representing the intention of the user is expanded. Based on the target search text and the user context information, a target search language chain may be constructed.

Further, according to the target search language chain, the target search language chain can be input into a similarity processing model for similarity calculation, the similarity processing model is a model capable of comparing similarity between two texts, and the similarity degree of the two texts can be judged by calculating semantic or structural similarity between the two texts, so that output of the similarity processing model, namely a first candidate item set, is obtained. At this time, the target search language chain and the first candidate item set can be subjected to item expansion so as to enrich search dependent data, and according to the target search language chain and the first candidate item set, search prompt information can be obtained, so that the search intention of a user can be better clarified, and a foundation is provided for further obtaining more accurate search results.

Further, the search prompt information is input into a large language model for optimization, the large language model can learn from the search prompt information, and background information and preference of a user are effectively captured, namely, preference of the user on commodities in a certain field is summarized according to the information. Then, a relationship between the product attributes and the user preferences may be established, and the recommendation system provides the generated first candidate set to the large language model once the large language model has knowledge of the user preferences. The large language model may further filter and rank the candidate sets according to the user's preferences, ensuring that the user is presented with a smaller, more relevant second candidate set of items from which the user may obtain more accurate target search results.

According to the embodiment, a target search language chain is obtained by acquiring a target search text and user context information and according to the target search text and the user context information; inputting the target search language chain into a similarity processing model to perform similarity calculation to obtain a first candidate item set output by the similarity processing model; obtaining search prompt information according to the target search language chain and the first candidate item set; inputting the search prompt information into the large language model to obtain a second candidate item set output by the large language model; and obtaining target search results according to the second candidate item set. The application can improve the searching accuracy of the searching system in the small sample field and the cold start field.

In some embodiments, before inputting the target search language chain into the similarity processing model for similarity calculation, the method comprises: acquiring a sample search text, and a plurality of positive sample information and a plurality of negative sample information corresponding to the sample search text; acquiring a first embedded vector according to the sample search text; acquiring a plurality of second embedded vectors according to the positive sample information and a plurality of third embedded vectors according to the negative sample information; model training is carried out according to the first embedded vector, the plurality of second embedded vectors and the plurality of third embedded vectors, and a preprocessing model is obtained; and determining target field parameters according to the user context information, and performing fine adjustment on the preprocessing model by utilizing the target field parameters to obtain a similarity processing model.

Specifically, multiple sets of sample search text, and multiple positive and negative sample information corresponding to each sample search text, may be collected before inputting the target search language chain into the similarity processing model for similarity calculation. Positive sample information is search results or related content related to the sample search text, while negative sample information is content not related to the sample search text.

The sample search text is preprocessed to obtain a first embedded vector. This vector is a representation of the sample search text, a fixed length vector generated by encoding or embedding the text. The system may obtain a plurality of second embedded vectors and a plurality of third embedded vectors by sample searching the text for a corresponding plurality of positive and negative sample information. These vectors are representations of related and unrelated content that may be generated by encoding or embedding.

Further, model training is performed using the first, second, and third embedded vectors. The goal of training is to distinguish between positive and negative samples by these vectors. During training, the system generates an appropriate model based on the similarity or difference between vectors, which can effectively distinguish between positive and negative samples when considering the similarity calculation.

Further, based on the user context information and domain-specific requirements of the target search, target domain parameters may be determined, which may include domain-specific weights, thresholds, or other relevant parameters. And then, fine-tuning the preprocessing model by using the target field parameters to obtain a final similarity processing model so as to adapt to the processing requirements of the similarity model in the feature field.

According to the embodiment, the sample search text is obtained, and a plurality of positive sample information and a plurality of negative sample information corresponding to the sample search text are obtained; acquiring a first embedded vector according to the sample search text; acquiring a plurality of second embedded vectors according to the positive sample information and a plurality of third embedded vectors according to the negative sample information; model training is carried out according to the first embedded vector, the plurality of second embedded vectors and the plurality of third embedded vectors, and a preprocessing model is obtained; and determining target field parameters according to the user context information, and performing fine adjustment on the preprocessing model by utilizing the target field parameters to obtain a similarity processing model, so that the quality of search results can be improved, and the correlation of field search can be better improved.

Further, in some embodiments, model training is performed based on the first embedded vector, the second plurality of embedded vectors, and the third plurality of embedded vectors to obtain a pre-processing model, comprising: calculating a first similarity of the first embedded vector and the plurality of second embedded vectors, and calculating a second similarity of the first embedded vector and the plurality of third embedded vectors; and constructing a model loss function based on the first similarity and the second similarity, and training the pretreatment model according to the model loss function to obtain the pretreatment model.

Specifically, from a given first embedded vector and a plurality of second embedded vectors, the similarity between the first embedded vector and each of the second embedded vectors may be calculated. The similarity may use different calculation methods, such as cosine similarity or euclidean distance. Similarly, from a given first embedded vector and a plurality of third embedded vectors, a similarity between the first embedded vector and each of the third embedded vectors may be calculated. Also, different similarity calculation methods may be used to measure the degree of similarity between two vectors.

Further, a loss function of the model may be constructed based on the first similarity and the second similarity. The goal of the penalty function is to make the similarity between the first embedded vector and the second embedded vector higher and the similarity between the third embedded vector lower.

It can be understood that when the loss function is used for training the information retrieval model, the cross entropy loss function or the triplet loss function can be used for training the model, and model parameters can be updated through optimization algorithms such as gradient descent, so that the loss function is reduced as much as possible, the learning capacity of the information retrieval model is optimized, and an accurate output result is obtained.

Additionally, in some embodiments, the similarity processing model includes a similarity processing network and a ranking network; inputting the target search language chain into a similarity processing model to perform similarity calculation to obtain a first candidate item set output by the similarity processing model, wherein the method comprises the following steps: performing similarity calculation processing on the target search language chain by using a similarity processing network to obtain a plurality of initial search results, wherein the similarity between the plurality of initial search results and the target search language chain is larger than a preset value; and sequencing the plurality of initial search results by utilizing a sequencing network to obtain a first candidate item set.

Specifically, the similarity processing model includes a similarity processing network and a ranking network, the similarity processing network can measure similarity between the target search language chain and an existing sample search. And filtering out a plurality of initial search results with similarity larger than a preset value by calculating the similarity score of the target search language chain and the sample search and comparing the similarity score with a preset similarity threshold.

Further, ranking the plurality of initial search results using a ranking network may generate a first set of candidate items. The ranking network may rank the initial search results according to relevance or other criteria according to a defined ranking algorithm or a learned ranking model. This process may rank search results that are more highly similar and relatively more relevant to the target search language chain in front, forming a first set of candidate terms.

The embodiment can place the search result which is more relevant to the target search language chain and has higher similarity in front through similarity calculation and sorting operation, and provides reference and selection for a user. The method can improve the accuracy and the relevance of search and provide more accurate search results in the fields of cold start and small samples.

In some embodiments, obtaining search hint information from a target search language chain and a first set of candidate terms includes: inputting the target search language chain and the first candidate item set into a large language model for expansion to obtain first expansion information corresponding to the target search language chain and second expansion information corresponding to the first candidate item set; and inputting the first expansion information and the second expansion information into a construction prompt network to obtain search prompt information.

Specifically, the target search language chain and the first candidate item set are used as inputs, and expansion processing is performed through a large language model. The large language model may generate more content related to the target search language chain based on the input. The process can obtain the first expansion information corresponding to the target search language chain, wherein the information contains other search keywords, phrases or related subjects related to target search, and the defect of insufficient small sample data can be overcome. At the same time, second expansion information corresponding to the first candidate item set can be obtained, and the information can comprise more search results or related contents related to the first candidate item.

Further, the first extension information and the second extension information are input to the construction hint network. The build prompting network may employ a predefined network structure or a trained model that can integrate and process such information and generate the final search prompting information. These hints may be related search keywords, related topics, or other content that may be relevant to the target search.

According to the embodiment, the target search language chain and the first candidate item set are input into the large language model, the generated expansion information can be used for guiding search, meanwhile, the target search language chain and the first candidate item set are expanded and the search prompt information is generated, so that the problems of small samples and cold start can be effectively solved, and the accuracy and the comprehensiveness of search are improved.

In addition, in some embodiments, the search hint information is input to the large language model to obtain a second candidate item set output by the large language model, including: inputting the search prompt information into a large language model to obtain a plurality of candidate targets; inputting the target search language chain into a large language model to capture preference information, and obtaining user preference information; and filtering the plurality of candidate targets according to the user preference information by using the large language model to obtain a second candidate item set.

Specifically, the search prompt information is input into a large language model, and a plurality of candidate targets can be acquired. By inputting search prompt information, the large language model can combine the language model, the context understanding and the semantic understanding in the large language model to generate a plurality of candidate targets related to target search.

Further, a target search language chain is input into the large language model to capture preference information of the user. By entering a target search language chain, the large language model can understand the user's search history, interests, and preferences. In this way, the model may better understand the needs of the user and perform subsequent filtering and sorting operations accordingly.

Further, the plurality of candidate targets are filtered according to preference information of the user by using the large language model to obtain a second candidate item set. By considering the user's preference information, the large language model may evaluate and rank multiple candidate targets, thereby picking targets that are more relevant to the user's interests and needs. Candidate targets that do not match the user's preferences may be filtered to provide a second set of candidate items that more closely match the user's needs.

According to the embodiment, the search prompt information is input into the large language model, the generated candidate targets are utilized and filtered by combining with the user preference information, so that a second candidate item set which is more matched with the user requirements is obtained, and the individuation degree and the accuracy of the search are further improved.

Additionally, in some embodiments, after obtaining the target search result according to the second candidate item set, further comprising: displaying the target search result through a visual interface, and receiving interactive operation aiming at the target search result; the target search results are adjusted based on the interaction operation.

Specifically, the second candidate item set may be presented to the user as target search results through a visual interface. May be a search results page on which the user can see a list of items or cards associated with their search needs. The search results may be ranked according to relevance, time, category, etc., to facilitate user review.

In addition, the user can also perform interactive operations such as clicking, scrolling, screening, collecting and the like through the visual interface. The user may click on a search result to view more detailed information or browse more search results by scrolling or paging through pages. The user may also use filtering options to narrow down the scope of the search results or to collect some of the search results for later reference.

And adjusting the search result according to the interactive operation of the user. For example, when a user clicks on a search result, the user may be recommended related items or similar items. When the user uses the filtering option, the search results may be re-filtered and items meeting the filtering criteria presented. The ranking of the search results or recommending more relevant content may be dynamically adjusted based on the user's interaction behavior.

According to the embodiment, the interactive operation of the user is received through the visual interface, the search result can be adjusted according to the interactive behavior of the user, and more accurate and personalized search experience is provided.

Any combination of the above optional solutions may be adopted to form an optional embodiment of the present application, which is not described herein.

It should be understood that the sequence number of each step in the above embodiment does not mean the sequence of execution sequence, and the execution sequence of each process should be determined by the function and the internal logic of each process, and should not be construed as limiting the process in the embodiment of the present application.

Fig. 2 is a schematic diagram of a target searching apparatus according to an embodiment of the present application. As shown in fig. 2, the apparatus includes:

an obtaining module 201 configured to obtain a target search text and user context information, and obtain a target search language chain according to the target search text and the user context information;

The similarity processing module 202 is configured to input the target search language chain into the similarity processing model to perform similarity calculation, so as to obtain a first candidate item set output by the similarity processing model;

The prompt module 203 is configured to obtain search prompt information according to the target search language chain and the first candidate item set;

The filtering module 204 is configured to input the search prompt information into the large language model to obtain a second candidate item set output by the large language model;

the search module 205 is configured to obtain a target search result according to the second candidate item set, wherein the target search result comprises a target search object and description information aiming at the target search object.

In some embodiments, the similarity processing module 202 is further configured to obtain a sample search text, and a plurality of positive sample information and a plurality of negative sample information corresponding to the sample search text; acquiring a first embedded vector according to the sample search text; acquiring a plurality of second embedded vectors according to the positive sample information and a plurality of third embedded vectors according to the negative sample information; model training is carried out according to the first embedded vector, the plurality of second embedded vectors and the plurality of third embedded vectors, and a preprocessing model is obtained; and determining target field parameters according to the user context information, and performing fine adjustment on the preprocessing model by utilizing the target field parameters to obtain a similarity processing model.

In some embodiments, the similarity processing module 202 is further configured to calculate a first similarity of the first embedded vector and the plurality of second embedded vectors, and calculate a second similarity of the first embedded vector and the plurality of third embedded vectors; and constructing a model loss function based on the first similarity and the second similarity, and training the pretreatment model according to the model loss function to obtain the pretreatment model.

In some embodiments, the similarity processing model includes a similarity processing network and a ranking network, and the similarity processing module 202 is further configured to perform similarity calculation processing on the target search language chain by using the similarity processing network to obtain a plurality of initial search results, where the similarity between the plurality of initial search results and the target search language chain is greater than a preset value; and sequencing the plurality of initial search results by utilizing a sequencing network to obtain a first candidate item set.

In some embodiments, the prompt module 203 is further configured to input the target search language chain and the first candidate item set into the large language model for expansion, so as to obtain first expansion information corresponding to the target search language chain and second expansion information corresponding to the first candidate item set; and inputting the first expansion information and the second expansion information into a construction prompt network to obtain search prompt information.

In some embodiments, the filtering module 204 is further configured to input the target search language chain into the large language model for capturing preference information, so as to obtain user preference information; inputting the search prompt information into a large language model to obtain a plurality of candidate targets; and filtering the plurality of candidate targets according to the user preference information by using the large language model to obtain a second candidate item set.

In some embodiments, the screening module 204 is further configured to display the target search result through a visual interface and receive an interaction with respect to the target search result; the target search results are adjusted based on the interaction operation.

The device provided by the embodiment of the application can realize all the method steps of the method embodiment and achieve the same technical effects, and is not described herein.

Fig. 3 is a schematic diagram of an electronic device 3 according to an embodiment of the present application. As shown in fig. 3, the electronic apparatus 3 of this embodiment includes: a processor 301, a memory 302 and a computer program 303 stored in the memory 302 and executable on the processor 301. The steps of the various method embodiments described above are implemented when the processor 301 executes the computer program 303. Or the processor 301 when executing the computer program 303 performs the functions of the modules/units in the above-described device embodiments.

The electronic device 3 may be an electronic device such as a desktop computer, a notebook computer, a palm computer, or a cloud server. The electronic device 3 may include, but is not limited to, a processor 301 and a memory 302. It will be appreciated by those skilled in the art that fig. 3 is merely an example of the electronic device 3 and is not limiting of the electronic device 3 and may include more or fewer components than shown, or different components.

The Processor 301 may be a central processing unit (Central Processing Unit, CPU) or other general purpose Processor, digital signal Processor (DIGITAL SIGNAL Processor, DSP), application SPECIFIC INTEGRATED Circuit (ASIC), field-Programmable gate array (Field-Programmable GATE ARRAY, FPGA) or other Programmable logic device, discrete gate or transistor logic device, discrete hardware components, or the like.

The memory 302 may be an internal storage unit of the electronic device 3, for example, a hard disk or a memory of the electronic device 3. The memory 302 may also be an external storage device of the electronic device 3, for example, a plug-in hard disk provided on the electronic device 3, a smart memory card (SMART MEDIA CARD, SMC), a Secure Digital (SD) card, a flash memory card (FLASH CARD), or the like. The memory 302 may also include both internal storage units and external storage devices of the electronic device 3. The memory 302 is used to store computer programs and other programs and data required by the electronic device.

It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-described division of the functional units and modules is illustrated, and in practical application, the above-described functional distribution may be performed by different functional units and modules according to needs, i.e. the internal structure of the apparatus is divided into different functional units or modules to perform all or part of the above-described functions. The functional units and modules in the embodiment may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit, where the integrated units may be implemented in a form of hardware or a form of a software functional unit.

The integrated modules/units may be stored in a readable storage medium if implemented in the form of software functional units and sold or used as stand-alone products. Based on such understanding, the present application may implement all or part of the flow of the method of the above embodiment, or may be implemented by a computer program to instruct related hardware, and the computer program may be stored in a readable storage medium, where the computer program may implement the steps of the method embodiments described above when executed by a processor. The computer program may comprise computer program code, which may be in source code form, object code form, executable file or in some intermediate form, etc. The readable storage medium may include: any entity or device capable of carrying computer program code, a recording medium, a U disk, a removable hard disk, a magnetic disk, an optical disk, a computer Memory, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), an electrical carrier signal, a telecommunications signal, a software distribution medium, and so forth.

The above embodiments are only for illustrating the technical solution of the present application, and are not limiting; although the application has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present application, and are intended to be included in the scope of the present application.

Claims (10)

1. A target search method, comprising:

Acquiring a target search text and user context information, and acquiring a target search language chain according to the target search text and the user context information;

inputting the target search language chain into a similarity processing model to perform similarity calculation to obtain a first candidate item set output by the similarity processing model;

obtaining search prompt information according to the target search language chain and the first candidate item set;

inputting the search prompt information into a large language model to obtain a second candidate item set output by the large language model;

And obtaining target search results according to the second candidate item set, wherein the target search results comprise target searches and description information aiming at the target searches.

2. The method of claim 1, wherein before inputting the target search language chain into a similarity processing model for similarity calculation, comprising:

Acquiring a sample search text, and a plurality of pieces of positive sample information and a plurality of pieces of negative sample information corresponding to the sample search text;

Searching text according to the sample to obtain a first embedded vector;

Acquiring a plurality of second embedded vectors according to the positive sample information and a plurality of third embedded vectors according to the negative sample information;

Model training is carried out according to the first embedded vector, the plurality of second embedded vectors and the plurality of third embedded vectors, so as to obtain a preprocessing model;

And determining target domain parameters according to the user context information, and performing fine adjustment on the preprocessing model by utilizing the target domain parameters to obtain the similarity processing model.

3. The method of claim 2, wherein the model training based on the first embedded vector, the plurality of second embedded vectors, and the plurality of third embedded vectors, results in a pre-processing model, comprising:

Calculating a first similarity of the first embedded vector and the plurality of second embedded vectors, and calculating a second similarity of the first embedded vector and the plurality of third embedded vectors;

And constructing a model loss function based on the first similarity and the second similarity, and training the pretreatment model according to the model loss function to obtain the pretreatment model.

4. The method of claim 1, wherein the similarity processing model comprises a similarity processing network and a ranking network;

inputting the target search language chain into a similarity processing model to perform similarity calculation, and obtaining a first candidate item set output by the similarity processing model, wherein the method comprises the following steps:

Performing similarity calculation processing on the target search language chain by using the similarity processing network to obtain a plurality of initial search results, wherein the similarity between the plurality of initial search results and the target search language chain is larger than a preset value;

And sequencing the plurality of initial search results by using the sequencing network to obtain a first candidate item set.

5. The method of claim 1, wherein the obtaining search hint information from the target search language chain and the first set of candidate items comprises:

inputting the target search language chain and the first candidate item set into a large language model for expansion to obtain first expansion information corresponding to the target search language chain and second expansion information corresponding to the first candidate item set;

and inputting the first extension information and the second extension information into a construction prompt network to obtain the search prompt information.

6. The method of claim 1, wherein inputting the search hint information into the large language model to obtain a second set of candidate items output by the large language model comprises:

Inputting the search prompt information into the large language model to obtain a plurality of candidate targets;

Inputting the target search language chain into the large language model to capture preference information, and obtaining user preference information;

And filtering the plurality of candidate targets according to the user preference information by using the large language model to obtain the second candidate item set.

7. The method of claim 1, wherein after obtaining the target search result from the second candidate item set, further comprising:

Displaying the target search result through a visual interface, and receiving interactive operation aiming at the target search result;

and adjusting the target search result based on the interaction operation.

8. A target search apparatus, comprising:

The acquisition module is configured to acquire target search text and user context information, and acquire a target search language chain according to the target search text and the user context information;

The similarity processing module is configured to input the target search language chain into a similarity processing model to perform similarity calculation, so as to obtain a first candidate item set output by the similarity processing model;

The prompt module is configured to obtain search prompt information according to the target search language chain and the first candidate item set;

The screening module is configured to input the search prompt information into a large language model to obtain a second candidate item set output by the large language model;

and the searching module is configured to obtain target search results according to the second candidate item set, wherein the target search results comprise target searches and description information aiming at the target searches.

9. An electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the steps of the method according to any of claims 1 to 7 when the computer program is executed.

10. A readable storage medium storing a computer program, characterized in that the computer program when executed by a processor implements the steps of the method according to any one of claims 1 to 7.