Search word analysis model data processing method and device and computer equipment

Abstract

The application relates to a search word analysis model data processing method and device, computer equipment, a storage medium and a computer program product. The method comprises the following steps: based on the correlation between different data in search log data, search scene training data is constructed; a large language model is subjected to model training processing to obtain a search scene large language model; through a context learning mode, task scene training data of a search word analysis task is constructed in a prompt word mode; based on the task scene training data, the search scene large language model is subjected to supervised fine-tuning training to obtain a search word analysis model. Through the correlation between data in search log data, the large language model is trained, and then the search word analysis task is unified into a generative task through the construction of task scene training data to complete fine-tuning training and obtain the search word analysis model, so that the analysis effect and analysis accuracy of the obtained search word analysis model in search word analysis are effectively ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computers, in particular to a search word analysis model data processing method and device, computer equipment, a storage medium and a computer program product. BACKGROUND

[0002] With the development of computer technology and nature language processing (NLP) technology, query word analysis technology has emerged. Search engine query word analysis refers to the process of natural language processing, semantic understanding, intent recognition and other technical processing on the query word input by the user in the search engine, so as to more accurately understand the user's demand, mine potential information value, and provide more relevant and high-quality search results for the user. Search engine search word analysis is a key link in search engine technology, which directly affects the performance and user experience of the search engine.

[0003] At present, the search word can be comprehensively analyzed by a deep neural network, but this way cannot guarantee the analysis effect of the search word. SUMMARY

[0004] Therefore, it is necessary to provide a search word analysis model data processing method, device, computer equipment, computer readable storage medium and computer program product capable of effectively improving the analysis effect of the search word.

[0005] In a first aspect, the present application provides a search word analysis model data processing method, comprising:

[0006] Based on the relevance between different data in the search log data, search scene training data is constructed;

[0007] The pre-trained large language model is retrained by the search scene training data to obtain a search scene large language model;

[0008] Task scene training data of the search word analysis task is constructed by a context learning method;

[0009] Based on the task scene training data, the search scene large language model is supervised and fine-tuned to obtain a search word analysis model.

[0010] In a second aspect, the present application further provides a search word analysis model data processing device, comprising:

[0011] The association analysis module is configured to construct search scene training data based on the relevance between different data in the search log data;

[0012] a model training module configured to perform model retraining on a pre-trained large language model based on the search scenario training data, to obtain a search scenario large language model;

[0013] a data construction module configured to construct task scenario training data for a search term analysis task in a context learning manner;

[0014] a model fine-tuning module configured to perform supervised fine-tuning training on the search scenario large language model based on the task scenario training data, to obtain a search term analysis model.

[0015] In a third aspect, the present application further provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0016] constructing search scenario training data based on the relevance between different data in search log data;

[0017] performing model retraining on a pre-trained large language model based on the search scenario training data, to obtain a search scenario large language model;

[0018] constructing task scenario training data for a search term analysis task in a context learning manner;

[0019] performing supervised fine-tuning training on the search scenario large language model based on the task scenario training data, to obtain a search term analysis model.

[0020] In a fourth aspect, the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the following steps:

[0021] constructing search scenario training data based on the relevance between different data in search log data;

[0022] performing model retraining on a pre-trained large language model based on the search scenario training data, to obtain a search scenario large language model;

[0023] constructing task scenario training data for a search term analysis task in a context learning manner;

[0024] performing supervised fine-tuning training on the search scenario large language model based on the task scenario training data, to obtain a search term analysis model.

[0025] In a fifth aspect, the present application further provides a computer program product, comprising a computer program, and the computer program is executed by a processor to implement the following steps:

[0026] constructing search scenario training data based on the relevance between different data in search log data;

[0027] retrain the pre-trained large language model through the search scenario training data to obtain a search scenario large language model;

[0028] construct task scenario training data of the search term analysis task through a context learning manner;

[0029] based on the task scenario training data, perform supervised fine-tuning training on the search scenario large language model to obtain a search term analysis model.

[0030] The search term analysis model data processing method, device, computer device, storage medium and computer program product can first construct search scenario training data based on the relevance between different data in search log data, and then retrain the pre-trained large language model through the search scenario training data to obtain a search scenario large language model. The training of the search scenario large language model can be completed by using real search log data to construct a corpus, which can better understand the query intention of the user in the search engine. Then, the task scenario training data of the search term analysis task is constructed through a context learning manner. The context learning manner can effectively construct a large amount of model training data associated with the search term analysis task. Finally, the search scenario large language model is supervised and fine-tuned based on the task scenario training data to obtain a search term analysis model. Thus, according to the specific search term analysis task, the search scenario large language model is adjusted and optimized in a fine-tuning manner, which can effectively ensure the processing performance of the search scenario large language model when processing the search term analysis task. The present application completes the training of the large language model based on the relevance of the data in the search log data to obtain a search field large language model, and then fine-tunes the search scenario large language model by constructing task scenario training data and unifying the search term analysis task into a generative task to obtain a search term analysis model, which effectively ensures the analysis effect and accuracy of the obtained search term analysis model in search term analysis. BRIEF DESCRIPTION OF DRAWINGS

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the embodiments or related art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0032] Figure 1 An application environment diagram of the search term analysis model data processing method in an embodiment;

[0033] Figure 2A flowchart of a search term analysis model data processing method in one embodiment;

[0034] Figure 3 A diagram of click content correlation analysis in one embodiment;

[0035] Figure 4 A diagram of search scenario training data constructed based on click content correlation analysis in one embodiment;

[0036] Figure 5 A diagram of search session correlation analysis in one embodiment;

[0037] Figure 6 A diagram of search scenario training data constructed based on search session correlation analysis in one embodiment;

[0038] Figure 7 A diagram of instruction part text in one embodiment;

[0039] Figure 8 A diagram of example part text in one embodiment;

[0040] Figure 9 A diagram of prediction part text in one embodiment;

[0041] Figure 10 A diagram of complete task scenario training data example in one embodiment;

[0042] Figure 11 A diagram of single task training sample in one embodiment;

[0043] Figure 12 A diagram of multi-task training sample in one embodiment;

[0044] Figure 13 A diagram of full task training sample in one embodiment;

[0045] Figure 14 A flowchart of a search term analysis model data processing method in another embodiment;

[0046] Figure 15 A structural block diagram of a search term analysis model data processing apparatus in one embodiment;

[0047] Figure 16 An internal structure diagram of a computer device in one embodiment. DETAILED DESCRIPTION

[0048] In order to make the purposes, technical solutions and advantages of the present application clearer, further detailed descriptions will be given below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and not used to limit the present application.

[0049] The present application relates to the field of artificial intelligence. Artificial intelligence is the theory, method, technology and application system for using digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive environment, acquire knowledge and use knowledge to obtain optimal results. In other words, artificial intelligence is a comprehensive technology of computer science, which attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines, so that machines have the functions of perception, reasoning and decision-making.

[0050] Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, both hardware and software technologies. Artificial intelligence basic technologies generally include technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, pre-training model technology, operation / interaction system, mechatronics, etc. Among them, the pre-training model is also called large model or basic model, which can be widely applied to downstream tasks in various directions of artificial intelligence after fine-tuning. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning, etc. The present application specifically relates to natural language processing technology and machine learning (Machine Learning, ML) technology in artificial intelligence.

[0051] Machine learning is a multi-disciplinary subject, involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, etc. It is a subject that studies how computers simulate or implement human learning behavior to acquire new knowledge or skills, and reorganize existing knowledge structure to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental approach to making computers intelligent, and its applications are widespread in various fields of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and inductive learning, etc. Pre-training model is the latest development of deep learning, which integrates the above technologies.

[0052] With the research and progress of artificial intelligence technology, artificial intelligence technology has been researched and applied in many fields, such as common smart home, smart wearable device, virtual assistant, smart speaker, smart marketing, unmanned vehicle, autonomous driving, unmanned aerial vehicle, digital twin, virtual human, robot, artificial intelligence generated content (AIGC), dialogue interaction, smart medical treatment, smart customer service, game AI, etc. It is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0053] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0054] The search term analysis model data processing method provided by the embodiments of the present application can be applied in an application environment as shown in Figure 1 . Wherein, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data required to be processed by the server 104. The data storage system can be integrated on the server 104, or placed on the cloud or other network servers. When the search analysis and operation personnel of the terminal 102 want to build a search term analysis model, and then realize the analysis and processing of the search term, a model building request can be submitted to the server 104, and the corresponding search log data is provided to build the analysis model. The server 104 will first build the search scene training data based on the correlation between different data in the search log data submitted by the terminal 102 after receiving the model building request; retrain the pre-trained large language model through the search scene training data to obtain the search scene large language model; build the task scene training data of the search term analysis task through the context learning method; based on the task scene training data, the search scene large language model is supervised and fine-tuned to obtain the search term analysis model. The terminal 102 can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things device can be a smart speaker, a smart television, a smart air conditioner, a smart vehicle device, etc. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, etc. The server 104 can be implemented by an independent server or a server cluster composed of multiple servers.

[0055] In an exemplary embodiment, as shown in Figure 2 , a search term analysis model data processing method is provided. Taking the server 104 in Figure 1 as an example, the method includes the following steps 201 to 207. Wherein:

[0056] In step 201, search scene training data is constructed based on the correlation between different data in the search log data.

[0057] The search log data refers to the record data generated by the search object during the data search process through the search engine. The search log data can record the search words (query) input by the target object, and can also record the time information related to the search, the search results obtained for the input search words, and the operation information of the target object on the search results, etc. The correlation between different data is mainly analyzed by the search log data to analyze the search query intention of the target object, so the correlation reflects the search intention of the user. For example, there is a correlation between the search results clicked by the target object for the input search words, and there is also a correlation between the search words input by the target object before and after. These correlations can be used to deeply mine the search query intention of the target object, thereby improving the accuracy of search word analysis. The search scene training data is used for model training data for query intention analysis, which can train a general large language model, so that the trained general large language model can adapt to the search scene and understand the search intention of the input search words of the target object.

[0058] For example, when the search analysis and operation personnel of the terminal 102 side hopes to construct a search word analysis model suitable for search word analysis, the corresponding model construction request can be submitted to the server 104, and the corresponding search log data is provided as a data reference for the search scene. The server 104 can construct search scene training data based on the search log data in the search scene, and then continue to train based on the general large language model to obtain a search domain large language model that understands the query intention of the input search words of the target object in the search engine and is suitable for the search scene. The search scene training data can be constructed by analyzing the correlation between different data in the search log data, which includes the correlation between the search results clicked by the target object for the input search words, and the correlation between the search words input by the target object before and after. Through the correlation analysis of the search log data, a large amount of search scene training data can be extracted. According to the search results clicked by the target object, it can be inferred that the user may be interested in which results. This can help the model better understand the correlation between the query and the related results, and also include all queries and browsing behaviors of the target object in a search session in the training data. Thus, it helps the model to understand the information demand changes of the target object in a session.

[0059] In step 203, the pre-trained large language model is retrained by the search scene training data to obtain a search scene large language model.

[0060] The large language model refers to a general large language model, i.e., a large language model applicable to any scene. By searching scene training data, the pre-trained large language model is retrained to obtain a large language model adapted to the search scene, i.e., a search scene large language model. Compared with the general large language model, the search scene large language model can better understand the query intention of the target object in the search engine.

[0061] Exemplarily, after obtaining the search scene training data, the optimization of the large language model can be completed through the search scene training data, the parameter update of the large language model is realized, and the search scene large language model is obtained. In one embodiment, the large language model can be implemented by a generative pre-trained transformer (GPT) model, which is a pre-trained language model based on the Transformer architecture. It is pre-trained on large-scale text data, and then fine-tuned on various NLP tasks to generate text, answer questions, and translate text. The core of the GPT model is the Transformer architecture, which is a deep learning model based on attention mechanism and can effectively process sequence data. In GPT, the input text is represented as a sequence, and the model generates the output text by processing the sequence. In the pre-training process, the GPT model learns the structure and patterns of language, which can be used for fine-tuning of various NLP tasks. In the fine-tuning process, the model adjusts and optimizes the pre-trained model according to the specific task to improve the performance of the model. The GPT model with a parameter scale of 7 billion can be selected as a general large language model in the present application, and then the general large language model is adjusted and optimized through the constructed search scene training data to obtain a search scene large language model.

[0062] In step 205, the task scene training data of the search word analysis task is constructed through the context learning method.

[0063] In Context Learning refers to a machine learning method that learns in a specific context environment. It takes into account the context environment of text, speech, image, video and other data, as well as the relationship between data and the influence of context information. In this method, the learning algorithm will use context information to improve the accuracy and effectiveness of prediction and classification. For example, in natural language processing, context learning can help machine learning algorithms better understand the meaning and relationship of words in a sentence. The search term analysis task is determined based on the search term analysis requirements, and different types of search term analysis requirements can be set up corresponding search term analysis tasks, and then the task scene training data is constructed based on the search term analysis task. For prompt words, in context learning, a "prompt" is given to the language model, which is a list of input-output pairs that describe a task. At the end of the prompt, there is a test input, and the language model is asked to predict the next token only based on the prompt. To correctly answer the following two prompts, the model needs to understand the demonstration examples of context learning to determine the input distribution, output distribution, input-output mapping, and format. Through context learning, a large amount of text data matching the search term analysis task can be constructed, which is the task scene training data.

[0064] Exemplarily, after obtaining the search scenario large language model, the search scenario large language model at this time can adapt to the search scenario and can analyze the search intent, but still cannot understand the specific search word analysis task requirement. Therefore, further training of the search scenario large language model needs to be completed. At this time, according to the specific search word analysis task requirement, the task scenario training data can be constructed in a prompt word manner through a context learning manner, and the task scenario training data constructed contains the requirement of input and output and examples of input and output, etc. In one of the embodiments, for search word analysis, generally contains key capabilities: intent recognition, that is, recognizing the query intent of the user, such as information retrieval, shopping, navigation, etc. Entity recognition, that is, recognizing entities in the query (such as names, places, brands, etc.), so as to more accurately provide relevant information. Keyword extraction, that is, extracting keywords from the user's query to better understand the topic and intent of the query. Query word expansion, that is, generating extended query words related to the original query words according to the query words input by the target object, so as to provide richer search results. Therefore, the search word analysis task can be constructed according to the requirement of these key capabilities, and the constructed search word analysis task includes domain classification, entity result and similar entity, keyword, functional word, word segmentation result and word weight scoring (0-4 points), timeliness, regional, query expansion, intent description, etc. On this basis, the task scenario training data constructed can contain task target, answer requirement and specific input and output part. For query word analysis task, the task target can be "you are a natural language understanding expert, please understand the search query word input by the user below", the answer requirement can be "analyze from multiple angles, including: domain classification, entity result and similar entity, keyword, functional word, word segmentation result and word weight scoring (0-4 points), timeliness, regional, query expansion, intent description, etc.", and then the input and output part is the specific query word and various analysis results for the query word.

[0065] In step 207, based on the task scenario training data, the search scenario large language model is supervised fine-tuning trained to obtain a search word analysis model.

[0066] Exemplarily, supervised fine-tuning (SFT) refers to using labeled data to adjust a pre-trained large language model to make it more suitable for a specific task. Usually, the pre-training of the large language model is unsupervised, but the fine-tuning process is often supervised. When supervised fine-tuning is performed, the model weights are adjusted according to the difference with the real label. Through this fine-tuning process, the model can capture the patterns and characteristics specific to a task in the labeled data. So that the model is more accurate and better adapts to a specific task.

[0067] Exemplarily, the task scene training data for constructing the search term analysis task in the form of a prompt word is labeled training data marked with analysis results corresponding to each search term analysis task. Therefore, after obtaining the task scene training data, the search scene large language model can be supervised and fine-tuned based on the task scene training data to obtain a search term analysis model, and the obtained search term analysis model can effectively meet the needs of various search term analysis tasks.

[0068] The search term analysis model data processing method first constructs search scene training data based on the relevance between different data in search log data, and then re-trains a pre-trained large language model based on the search scene training data to obtain a search scene large language model. The search scene large language model is trained by using real search log data to construct a corpus, which can better understand the query intent of the user in the search engine. Then, the task scene training data for the search term analysis task is constructed by context learning. The context learning can effectively construct a large amount of model training data associated with the search term analysis task. Finally, the search scene large language model is supervised and fine-tuned based on the task scene training data to obtain a search term analysis model. Thus, according to the specific search term analysis task, the search scene large language model is adjusted and optimized in a fine-tuning manner, which can effectively ensure the processing performance of the search scene large language model when processing the search term analysis task. The present application completes the training of the large language model to obtain a search field large language model based on the relevance of the data in the search log data, and then fine-tunes the search scene large language model by constructing task scene training data to unify the search term analysis task into a generative task, to obtain a search term analysis model, which effectively ensures the analysis effect and accuracy of the search term analysis model.

[0069] In an exemplary embodiment, step 201 includes: extracting input search term data in search log data; finding access content data determined based on the input search term data in the search log data; and constructing an association between the input search term data and the access content data to obtain search scene training data.

[0070] The input search term data is the text information input by the target object in the search box of the search engine interface. The access content data determined based on the input search term data refers to the information of the access interface clicked by the target object on the search result interface after the search result interface is fed back to the target object based on the search term input by the target object.

[0071] Exemplarily, for the correlation analysis process between different data in the search log data, the correlation between the input search words of the target object and the search results can be analyzed, and according to the search results clicked by the target object, it can be inferred that the target object is likely to be interested in which results. This can help the model better understand the correlation between the query and the relevant results. Therefore, when performing correlation analysis, the input search word data in the search log data can be extracted first, and then for each search word, the content data in the search log data is accessed to determine which search results the target object is interested in, so that the correlation between the input search word data and the accessed content data is established, and the corresponding search scene training data is obtained. In one of the embodiments, in order to simplify the data, when accessing the content data, the complete original access content data determined based on the input search word data can be searched in the search log data first; and then the original access content data is processed to extract key information to obtain simplified access content data. For example, the result title, abstract or keyword clicked by the target object can be taken as part of the training data. In one of the embodiments, as shown in Figure 3 , the query word input by the target object is "how to bake a cake", and the search result titles clicked by the target object include "simple cake baking skills for beginners", "step-by-step guide to baking a cake at home", and "delicious cake recipes you must try", at this time, the correlation between the three search result titles clicked by the target object and the query word can be established, and 3 search scene training data as shown in Figure 4 is created. In this embodiment, by identifying the correlation between the input search word data and the access content data, the search scene training data is created, which can effectively mine the search content interested by the target object, thereby improving the effect of search word analysis.

[0072] In one of the example embodiments, step 201 includes: based on the log record generation time in the search log data, the search log data is split to obtain search session log data; and based on the correlation between different data in the search session log data, scene model training data is constructed.

[0073] Among them, for the log record generation time of the search log data, each log record will contain the corresponding record time, which can record the time point of the target object query or the time point of the query result click. The search session log data is obtained by splitting the search log data, and the search data in the same search session is associated.

[0074] Exemplarily, for the correlation analysis process between different data in the search log data, in addition to analyzing the correlation between the input search term and the search result of the target object, all queries and browsing behaviors of the target object in a search session can also be included in the training data. Thus helping the model to understand the information needs change of the target object in a session. Therefore, when performing correlation analysis, the search log data can be split based on the log record generation time in the search log data to obtain search session log data. For example, search log data generated within the same time period can be taken as a search session log data according to a fixed time interval, and a log time threshold can also be set to take search log data generated within the log time threshold as correlated search session log data. Then, the scene model training data is constructed based on the correlation between different data in the search session log data. In one of the embodiments, for the analysis process of the data correlation in the search session log data, the keywords of the input search term data in the search session log data can be extracted. Then the keywords are sorted based on the query time of the input search term data to obtain a keyword sorting result; finally, the keyword modification information is obtained according to the keyword sorting result, and the keyword modification information is taken as the scene model training data. In one of the embodiments, as shown in Figure 5 the target object successively queries the following keywords "weight loss tips", "healthy weight loss recipes", and "low-calorie recipes" in a search session, at this time, these information can be combined to generate a training data sample as shown in Figure 6 to represent the information needs change of the target object in a session. In this embodiment, by constructing search session data and then analyzing the correlation of the data in the search session data, the search scene training data is created, which can effectively mine the information needs change of the target object in a session, thereby improving the effect of search term analysis.

[0075] In one exemplary embodiment, step 205 includes: obtaining task scene training data examples of a search term analysis task constructed in a prompt word manner; performing context learning text prediction processing on the task scene training data examples based on the sample to generate a big data model, to obtain task scene training data.

[0076] The prompt mode corresponds to context learning, which refers to using a prompt containing an instruction part, an example part, and a prediction part as a task scene training data example. The data example can be constructed according to the style of the training data required by the search word analysis model. The sample generation big data model is used to understand the task scene training data example, and the context learning method is used to predict the missing data in the task scene training data example, so as to obtain the task scene training data for the search word analysis task. The sample generation big data model is also a large language model, which can be iterated based on the task scene training data example to generate a large amount of task scene training data to complete the subsequent search word analysis model training task.

[0077] Exemplarily, when a large amount of character model training data needs to be constructed, in order to improve the generation efficiency of the task scene training data and obtain a large amount of training data for training the search word analysis model. The task scene training data can be generated in the prompt mode based on context learning through the sample generation big data model. First, the task scene training data example of the search word analysis task constructed in the prompt mode can be obtained, which contains the contents of the instruction part, the example part, and the prediction part. The sample generation big data model can refer to the requirements of the instruction part, and predict the content of the prediction part by imitating the examples listed in the example part, so as to obtain the task scene training data. In specific embodiments, the generated task scene training data can be iterated through the sample generation big data model to construct more optimal prompt examples and task scene training data. In this embodiment, the sample generation big data model is used to predict the task scene training data example text, so as to generate a large amount of task scene training data, which can effectively guarantee the generation efficiency of the task scene training data, and further guarantee the analysis effect of the search word analysis model.

[0078] In an exemplary embodiment, obtaining the task scene training data example of the search word analysis task constructed in the prompt mode includes: constructing a task target and an answer requirement based on the task type of the search word analysis task to obtain an instruction part text; constructing an example part text based on the task background information and the expected prediction result of the search word analysis task, and constructing a prediction part text corresponding to the example part text; and splicing the instruction part text, the example part text, and the prediction part text to obtain the task scene training data example.

[0079] The instruction part provides the goal and requirement of the task, thereby limiting the sample generation big data model to answer the question according to the specified standard. The instruction part specifically includes two parts of task goal and answer requirement. The task goal refers to the goal to be achieved when the search term is analyzed, and the answer requirement is used to limit the sample generation big data model to answer the question according to the specified standard. In one embodiment, the content of the instruction part applied to the search term analysis can refer to the content shown in Figure 7 The example part text provides examples and instances related to the task, helping the big model to further understand the specific ability requirements of the task. In one task scene training data example, multiple example part texts can be provided for the big model to learn. A single example part is divided into an input part and an output part. The input part provides the background information of the task for the big model, and the output part is the expected prediction result of the big model. In one embodiment, the content of the example part applied to the search term analysis can refer to the content shown in Figure 8 The prediction part text refers to the real use case that the model needs to predict. It is the same in form as the example part text, but the output part is empty, and the answer is obtained by the model to predict. In one embodiment, the content of the prediction part applied to the search term analysis can refer to the content shown in Figure 9 And the complete task scene training data example obtained by splicing the three can refer to the content shown in Figure 10

[0080] ​Exemplarily, when constructing the task scene training data, the instruction part text can be constructed based on the task type of the search word analysis task to construct the task target and the answer requirement, so as to inform the sample generation big data model of the requirement of task processing and the requirement and format that the sample generation big data model should follow for text prediction processing. Then, the example part text is constructed based on the task background information and the expected prediction result of the search word analysis task, and the prediction part text corresponding to the example part text is constructed, and the sample generation big data model can refer to the example to understand the specific requirement of the task scene training data generation task. Meanwhile, the prediction part text corresponding to the example part text also needs to be constructed in the instance. When generating the task scene training data, the sample generation big data model can refer to the example part text, and based on the sample generation big data model, the context learning text prediction processing of the task scene training data example is performed to obtain the text prediction result of the task background information of the prediction part text in the task scene training data example; and the text prediction result is added to the prediction part text in the task scene training data example to obtain the task scene training data, so as to generate a large amount of task scene training data. The instruction part text, the example part text and the prediction part text are spliced to obtain the task scene training data example. In this embodiment, the instruction part text, the example part text and the prediction part text associated with the search word analysis task are constructed, and then spliced to form the task scene training data example, so that the sample generation big data model can refer to the requirement and example of the task scene training data example to perform text prediction, and a large amount of task scene training data is effectively generated.

[0081] In one embodiment, step 207 includes: simplifying the task scene training data to obtain a training data simplification result; and performing supervised fine-tuning training on the search scene large language model based on the training data simplification result to obtain a search word analysis model.

[0082] Exemplarily, the simplified processing refers to simplified processing of the prompt word mode task scene training data. The task scene training data in the complete prompt word mode contains the content of the instruction part text, the example part text and the prediction part text. The content of the example part text is similar to the processed prediction part text. Therefore, it is necessary to simplify the content. Therefore, in the simplified processing, the example part text in the task scene training data can be removed, and the instruction part text and the prediction part text are retained. Then, the two part texts are used as the training data simplified result to supervise the fine-tuning training of the search scene large language model to obtain the search word analysis model. In the embodiment, the training of the search word analysis model is realized by the simplified processing of the task scene training data. The overall length of the task scene training data is greatly shortened, the model inference efficiency of the training process can be improved, and the online service deployment cost is reduced.

[0083] In one embodiment, step 207 includes: based on the task type of the search word analysis task, performing split processing on the task scene training data to obtain single-task training samples corresponding to search word analysis tasks of different task types; performing random combination processing on the single-task training samples to obtain reconstructed multi-task training samples; based on the reconstructed multi-task training samples, performing supervised fine-tuning training on the search scene large language model to obtain the search word analysis model.

[0084] In one embodiment, step 207 includes: based on the task type of the search word analysis task, performing split processing on the task scene training data to obtain single-task training samples corresponding to search word analysis tasks of different task types; performing random combination processing on the single-task training samples to obtain reconstructed multi-task training samples; based on the reconstructed multi-task training samples, performing supervised fine-tuning training on the search scene large language model to obtain the search word analysis model.

[0085] Exemplarily, for the process of supervised fine-tuning training, the task scene training data can be split processed based on the task type of the search word analysis task to obtain single-task training samples corresponding to search word analysis tasks of different task types. The single-task training samples are randomly combined to obtain reconstructed multi-task training samples. Through the split and combination of the text data in the training samples, the random combination of the tasks is realized, and samples suitable for different task quantities are generated. Therefore, after the model training is completed, different task scenes can be self-adapted, and single or multiple tasks can be freely selected for prediction. The example of the single-task training sample of the domain classification can be referred to Figure 11The reconstruction multi-task training samples composed of the four tasks of domain classification, entity result, instruction classification, and instruction topic can refer to Figure 12 The full-task training samples combining all single-task training samples can refer to Figure 13 The corresponding model parameter size and parameter settings can refer to the table shown below.

[0086]

[0087] In this embodiment, the task type of the search word analysis task is analyzed to split and recombine the task scene training data, so as to realize the search word analysis model, so that the search word analysis model can automatically adapt to different search word analysis task scenes and improve the accuracy of search word analysis in these task scenes.

[0088] In one of the embodiments, the method further comprises: obtaining search word input information; performing search word analysis processing on the search word input information through the search word analysis model to obtain an intent recognition result, an entity word recognition result, a keyword extraction result, and a search word expansion result of the search word input information. Based on at least one of the intent recognition result, the entity word recognition result, the keyword extraction result, and the search word expansion result, auxiliary search processing is performed to obtain a target search result.

[0089] For example, after the training of the search word analysis model is completed, the corresponding search word analysis task can be completed through the search word analysis model obtained by training. The search word analysis task can include the content of the four fields of intent recognition result, entity word recognition result, keyword extraction result, and search word expansion result. The search word input information can be analyzed through the search word analysis model, and the intent recognition result, the entity word recognition result, the keyword extraction result, and the search word expansion result can be obtained through the output data of the model. The intent recognition result can identify the query intent of the search word input information, such as information retrieval, shopping, navigation, etc. The entity recognition result can identify the entity in the query (such as name, place name, brand, etc.), so as to more accurately provide related information. The keyword extraction result can extract keywords from the search word input information to better understand the theme and intent of the query. The search word expansion result is to generate an expanded search word related to the original search word according to the search word input information to provide more rich search results. Based on at least one of the intent recognition result, the entity word recognition result, the keyword extraction result, and the search word expansion result, auxiliary search processing is performed to obtain a target search result. In this embodiment, the search word input information is analyzed through the search word analysis model, and various types of search word analysis results can be obtained, thereby effectively improving the accuracy and efficiency of search word analysis.

[0090] The application also provides an application scenario, which is used to illustrate the search term analysis model data processing method described above. The search term analysis model data processing method specifically includes the following steps:

[0091] When an operator wants to provide more targeted operation services by analyzing the search terms of users in a search engine, the search term analysis model data processing method of the application can be used to construct a search term analysis model, and then the search term analysis processing can be completed through the obtained search term analysis model. Therefore, it is necessary to first construct a search term analysis model. In the model construction, first, the search scene training data can be constructed based on the correlation between different data in the search log data, and then the pre-trained large language model is retrained through the search scene training data to obtain a search scene large language model, so that the general large language model can be applied to the search scene.

[0092] For the correlation analysis process, first, the input search term data in the search log data can be extracted; in the search log data, the original access content data determined based on the input search term data is searched; the key information extraction processing is performed on the original access content data to obtain the access content data, and the correlation between the input search term data and the access content data is constructed to obtain the search scene training data. The second is to split the search log data based on the log record generation time to obtain the search session log data; the key words of the input search term data in the search session log data are extracted; the key words are sorted based on the query time of the input search term data to obtain the key word sorting result; the key word modification information is obtained according to the key word sorting result, and the key word modification information is used as the scene model training data. Then, the search scene large language model needs to be fine-tuned to adapt to the demand of search term analysis. At this time, the task target and answer requirement can be constructed based on the task type of the search term analysis task to obtain the instruction part text; the example part text is constructed based on the task background information and the expected prediction result of the search term analysis task, and the prediction part text corresponding to the example part text is constructed; the instruction part text, the example part text and the prediction part text are spliced to obtain the task scene training data example; the sample generation big data model is used to perform the context learning text prediction processing on the task scene training data example to obtain the text prediction result of the task background information of the prediction part text in the task scene training data example; the text prediction result is added to the prediction part text in the task scene training data example to obtain a large amount of task scene training data.

[0093] In order to further improve the training efficiency of the subsequent training, the task scene training data can also be simplified, and the task scene training data is split based on the task type of the search word analysis task to obtain single-task training samples corresponding to the search word analysis task of different task types; the single-task training samples are randomly combined to obtain reconstructed multi-task training samples; and the search scene large language model is supervised and fine-tuned based on the reconstructed multi-task training samples to obtain the search word analysis model. When the search word analysis is needed, the search word input information can be obtained; the search word input information is analyzed by the search word analysis model to obtain the intent recognition result, the entity word recognition result, the keyword extraction result and the search word expansion result for the search word input information, so as to realize the search word analysis. Through experimental measurement, as shown in Table 2, compared with the existing search word analysis model, the search word analysis model based on the large language model of the present application greatly reduces the model development cost, and the task accuracy is greatly improved.

[0094]

[0095] In one embodiment, the search word analysis model data processing method of the present application specifically includes Figure 14 as shown in the following table 2, compared with the existing search word analysis model, the search word analysis model based on the large language model of the present application greatly reduces the model development cost, and the task accuracy is greatly improved.

[0096] Step 1401, extract the input search word data in the search log data. Step 1403, in the search log data, find the original access content data determined based on the input search word data. Step 1405, perform key information extraction processing on the original access content data to obtain access content data. Step 1407, build the association relationship between the input search word data and the access content data to obtain search scene training data. Step 1409, retrain the pre-trained large language model through the search scene training data to obtain the search scene large language model. Step 1411, based on the task type of the search word analysis task, build the task target and the answer requirement to obtain the instruction part text. Step 1413, based on the task background information and the expected prediction result of the search word analysis task, build the example part text, and build the prediction part text corresponding to the example part text. Step 1415, splice the instruction part text, the example part text and the prediction part text to obtain the task scene training data example. Step 1417, based on the sample generation big data model, perform text prediction processing on the task scene training data example for context learning to obtain the text prediction result of the task background information of the prediction part text in the task scene training data example. Step 1419, add the text prediction result to the prediction part text in the task scene training data example to obtain the task scene training data. Step 1421, based on the task type of the search word analysis task, perform splitting processing on the task scene training data to obtain single-task training samples corresponding to search word analysis tasks of different task types. Step 1423, randomly combine the single-task training samples to obtain reconstructed multi-task training samples. Step 1425, based on the reconstructed multi-task training samples, perform supervised fine-tuning training on the search scene large language model to obtain the search word analysis model.

[0097] It should be understood that, although each step in the flowchart involved in the above embodiments is displayed in sequence according to the arrow indication, these steps are not necessarily executed in sequence according to the arrow indication. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other sequences. Moreover, at least part of the steps in the flowchart involved in the above embodiments can include multiple steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or steps or stages in other steps.

[0098] Based on the same inventive concept, the embodiments of the present application also provide a search word analysis model data processing device for implementing the search word analysis model data processing method described above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more search word analysis model data processing device embodiments provided below can refer to the limitations of the search word analysis model data processing method described above, which will not be repeated here.

[0099] In one exemplary embodiment, as shown in Figure 15 a search word analysis model data processing device is provided, comprising:

[0100] The association analysis module 1502 is configured to construct search scene training data based on the association between different data in the search log data.

[0101] The model training module 1504 is configured to perform model retraining on the pre-trained large language model through the search scene training data to obtain a search scene large language model.

[0102] The data construction module 1506 is configured to construct task scene training data of a search word analysis task through a context learning manner.

[0103] The model fine-tuning module 1508 is configured to perform supervised fine-tuning training on the search scene large language model based on the task scene training data to obtain a search word analysis model.

[0104] In one embodiment, the association analysis module 1502 is specifically configured to: extract input search word data in the search log data; find access content data determined based on the input search word data in the search log data; construct an association relationship between the input search word data and the access content data to obtain search scene training data.

[0105] In one embodiment, the association analysis module 1502 is specifically configured to: find original access content data determined based on the input search word data in the search log data; and perform key information extraction processing on the original access content data to obtain access content data.

[0106] In one embodiment, the association analysis module 1502 is specifically configured to: split the search log data based on the log record generation time in the search log data to obtain search session log data; and construct scene model training data based on the association between different data in the search session log data.

[0107] In an embodiment, the association analysis module 1502 is specifically configured to: extract a keyword of input search term data in the search session log data; perform sorting processing on the keyword based on a query time of the input search term data, to obtain a keyword sorting result; and obtain keyword modification information according to the keyword sorting result, and use the keyword modification information as scene model training data.

[0108] In an embodiment, the data construction module 1506 is specifically configured to: obtain a task scene training data example of a search term analysis task constructed in a prompt word manner; perform text prediction processing on the task scene training data example based on a sample generation big data model and context learning, to obtain task scene training data.

[0109] In an embodiment, the data construction module 1506 is specifically configured to: construct an instruction part text based on a task type of the search term analysis task and a requirement for an answer, construct an example part text based on task background information and an expected prediction result of the search term analysis task, and construct a prediction part text corresponding to the example part text; and splice the instruction part text, the example part text, and the prediction part text, to obtain the task scene training data example.

[0110] In an embodiment, the data construction module 1506 is specifically configured to: perform text prediction processing on the task scene training data example based on a sample generation big data model and context learning, to obtain a text prediction result of task background information of the prediction part text in the task scene training data example; and add the text prediction result to the prediction part text in the task scene training data example, to obtain the task scene training data.

[0111] In an embodiment, the model fine-tuning module 1508 is specifically configured to: perform simplification processing on the task scene training data, to obtain a training data simplification result; perform supervised fine-tuning training on the search scene large language model based on the training data simplification result, to obtain the search term analysis model.

[0112] In an embodiment, the model fine-tuning module 1508 is specifically configured to: perform splitting processing on the task scene training data based on a task type of the search term analysis task, to obtain single-task training samples corresponding to search term analysis tasks of different task types; perform random combination processing on the single-task training samples, to obtain reconstructed multi-task training samples; and perform supervised fine-tuning training on the search scene large language model based on the reconstructed multi-task training samples, to obtain the search term analysis model.

[0113] In one embodiment, a search analysis module is also included, which is used to: obtain search term input information; perform search term analysis on the search term input information through a search term analysis model to obtain intent recognition results, entity word recognition results, keyword extraction results and search term expansion results of the search term input information; perform auxiliary search processing based on at least one of the intent recognition results, entity word recognition results, keyword extraction results and search term expansion results to obtain target search results.

[0114] Each module in the above-mentioned search term analysis model data processing device can be implemented in whole or in part by software, hardware, or a combination thereof. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the corresponding operations of each of the above modules.

[0115] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 16 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store data related to the search term analysis model data processing process. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a search term analysis model data processing method is implemented.

[0116] Those skilled in the art will understand that Figure 16 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0117] In one embodiment, a computer device is further provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0118] In an embodiment, a computer readable storage medium storing a computer program is provided. The computer program, when executed by a processor, implements the steps of any of the above method embodiments.

[0119] In an embodiment, a computer program product or computer program including computer instructions stored in a computer readable storage medium is provided. A processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to cause the computer device to perform the steps of any of the above method embodiments.

[0120] It can be understood by those skilled in the art that all or part of the processes in the above embodiments can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above embodiments. Any reference to a memory, database or other medium used in the embodiments provided in the present application can include at least one of a non-volatile and volatile memory. The non-volatile memory can include a read-only memory (ROM), a magnetic tape, a floppy disk, a flash memory, an optical storage, a high-density embedded non-volatile memory, a resistive memory (ReRAM), a magnetoresistive random access memory (MRAM), a ferroelectric memory (FRAM), a phase change memory (PCM), a graphene memory, etc. The volatile memory can include a random access memory (RAM) or an external cache memory, etc. As an illustration but not limitation, the RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., and is not limited thereto. The processor involved in the embodiments provided in the present application can be a general processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., and is not limited thereto.

[0121] Any technical features in the above embodiments can be combined, and for the sake of brevity, not all possible combinations are described above, however, as long as the combinations of technical features do not have contradictions, they shall be considered within the scope of the present disclosure.

[0122] The above embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it shall not be understood as a limitation on the patent scope of the present application. It shall be pointed out that, for ordinary skilled persons in the art, several modifications and improvements can be made without departing from the concept of the present application, and these shall be within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A search term analysis model data processing method, characterized by, The method comprises: Based on the association between different data in the search log data, the search scene training data is constructed, and the association includes the association between the search results clicked by the target object for the input search word and the association between the search words input by the target object before and after; Through the search scene training data, the pre-trained large language model is retrained to obtain a search scene large language model, which is used to understand the query intention of the target object; Through the context learning mode, the task scene training data of the search word analysis task is constructed in the form of prompt words, and the search word analysis task includes intention recognition, entity recognition, keyword extraction and query word expansion; Based on the task scene training data, the search scene large language model is supervised and fine-tuned to obtain a search word analysis model; The task scene training data of the search word analysis task is constructed in the form of prompt words through the context learning mode, which comprises: Based on the task type of the search word analysis task, the task target and the answer requirement are constructed to obtain the instruction part text, the task target is the target to be achieved during the search word analysis, and the answer requirement is used to limit the sample generation big data model to answer the question according to the specified standard; Based on the task background information and the expected prediction result of the search word analysis task, the example part text is constructed, and the prediction part text corresponding to the example part text is constructed; The instruction part text, the example part text and the prediction part text are spliced to obtain the task scene training data example; Based on the sample generation big data model, the text prediction processing of the context learning of the task scene training data example is carried out to obtain the text prediction result of the task background information of the prediction part text in the task scene training data example; The text prediction result is added to the prediction part text in the task scene training data example to obtain the task scene training data.

2. The method of claim 1, wherein, The search scene training data is constructed based on the association between different data in the search log data, which comprises: Extract the input search word data in the search log data; In the search log data, find the access content data determined based on the input search word data; The association between the input search word data and the access content data is constructed to obtain the search scene training data.

3. The method of claim 2, wherein, The access content data determined based on the input search word data is found in the search log data, which comprises: In the search log data, find the original access content data determined based on the input search word data; The key information extraction processing is carried out on the original access content data to obtain the access content data.

4. The method of claim 1, wherein, The search scene training data is constructed based on the association between different data in the search log data, which comprises: Based on the log record generation time in the search log data, the search log data is split to obtain search session log data; Based on the association between different data in the search session log data, the scene model training data is constructed.

5. The method of claim 4, wherein, The constructing scene model training data based on the correlation between different data in the search session log data comprises: extracting keywords of input search term data in the search session log data; sorting the keywords based on the query time of the input search term data to obtain a keyword sorting result; obtaining keyword modification information according to the keyword sorting result, and taking the keyword modification information as scene model training data.

6. The method of claim 1, wherein, The supervision fine-tuning training of the search scene large language model based on the task scene training data comprises: simplifying the task scene training data to obtain a training data simplification result; based on the training data simplification result, the search scene large language model is supervised and fine-tuned to obtain a search term analysis model.

7. The method of claim 1, wherein, The supervision fine-tuning training of the search scene large language model based on the task scene training data comprises: based on the task type of the search term analysis task, the task scene training data is split to obtain single-task training samples corresponding to search term analysis tasks of different task types; randomly combining the single-task training samples to obtain reconstructed multi-task training samples; based on the reconstructed multi-task training samples, the search scene large language model is supervised and fine-tuned to obtain a search term analysis model.

8. The method according to any one of claims 1 to 7, characterized in that, The method further comprises: obtaining search term input information; performing search term analysis processing on the search term input information through the search term analysis model to obtain an intent recognition result, an entity word recognition result, a keyword extraction result, and a search term expansion result of the search term input information; based on at least one of the intent recognition result, the entity word recognition result, the keyword extraction result, and the search term expansion result, performing auxiliary search processing to obtain a target search result.

9. A search term analysis model data processing apparatus characterized by comprising: The device comprises: an association analysis module configured to construct search scene training data based on the correlation between different data in search log data, the correlation comprising the correlation between search results clicked by a target object for input search terms and the correlation between search terms input by the target object before and after; a model training module configured to retrain a pre-trained large language model based on the search scene training data to obtain a search scene large language model, the search scene large language model being configured to understand the query intent of a target object; a data construction module configured to construct task scene training data for a search term analysis task through a context learning method, the search term analysis task comprising intent recognition, entity recognition, keyword extraction, and query term expansion; a model fine-tuning module configured to supervise and fine-tune the search scene large language model based on the task scene training data to obtain a search term analysis model; The data construction module is specifically configured to: based on a task type of a search term analysis task, construct a task target and an answer requirement to obtain an instruction part text, the task target being a target to be achieved when analyzing a search term, and the answer requirement being used to limit a sample generation big data model to answer a question according to a specified standard; based on task background information and an expected prediction result of the search term analysis task, construct an example part text and a prediction part text corresponding to the example part text; splice the instruction part text, the example part text and the prediction part text to obtain a task scene training data example; based on a sample generation big data model, perform context learning text prediction processing on the task scene training data example to obtain a text prediction result of task background information of the prediction part text in the task scene training data example; and add the text prediction result to the prediction part text in the task scene training data example to obtain task scene training data.

10. The apparatus of claim 9, wherein, The correlation analysis module is specifically configured to: extract input search term data in search log data; in the search log data, find access content data determined based on the input search term data; and construct a correlation between the input search term data and the access content data to obtain search scene training data.

11. The apparatus of claim 10, wherein, The correlation analysis module is specifically configured to: in the search log data, find original access content data determined based on the input search term data; and perform key information extraction processing on the original access content data to obtain access content data.

12. The apparatus of claim 9, wherein, The correlation analysis module is specifically configured to: based on log record generation time in search log data, perform splitting processing on the search log data to obtain search session log data; and based on correlation between different data in the search session log data, construct scene model training data.

13. The apparatus of claim 12, wherein, The correlation analysis module is specifically configured to: extract keywords of input search term data in the search session log data; based on query time of the input search term data, perform sorting processing on the keywords to obtain a keyword sorting result; and based on the keyword sorting result, obtain keyword modification information, and use the keyword modification information as scene model training data.

14. The apparatus of claim 9, wherein, The model fine-tuning module is specifically configured to: perform simplification processing on the task scene training data to obtain a training data simplification result; based on the training data simplification result, perform supervised fine-tuning training on the search scene large language model to obtain a search term analysis model.

15. The apparatus of claim 9, wherein, The model fine-tuning module is specifically configured to: based on a task type of a search term analysis task, perform splitting processing on the task scene training data to obtain single-task training samples corresponding to search term analysis tasks of different task types; and perform random combination processing on the single-task training samples to obtain reconstructed multi-task training samples. Based on the reconstructed multi-task training samples, the search scene large language model is subjected to supervised fine-tuning training to obtain a search term analysis model.

16. The apparatus of any one of claims 9 to 15, wherein, The search analysis module is further configured to: acquire search term input information; perform search term analysis on the search term input information based on the search term analysis model to obtain an intent recognition result, an entity word recognition result, a keyword extraction result, and a search term expansion result of the search term input information; perform auxiliary search processing based on at least one of the intent recognition result, the entity word recognition result, the keyword extraction result, and the search term expansion result to obtain a target search result. 17.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-16. The processor, when executing the computer program, implements the steps of the method in any one of claims 1 to 8.

18. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by the processor, implements the steps of the method in any one of claims 1 to 8.

19. A computer program product comprising a computer program, characterized in that, The computer program, when executed by the processor, implements the steps of the method in any one of claims 1 to 8.

Images