Media content intelligent retrieval system and method based on large model

Abstract

The invention relates to the field related to artificial intelligence, and discloses an intelligent media content retrieval system and method based on a large language model, the large language model is used for training based on a large amount of corpus data, language laws and knowledge can be automatically learned, manual intervention is not needed, and therefore higher adaptability and generalization ability are achieved; and through technologies such as multi-language pre-training and the like, multiple languages are processed at the same time, cross-language information transmission and knowledge sharing can be realized, sentiment analysis of user problems is realized, real demands of the user are captured, problem intentions of the user are identified, and corresponding results are output according to different intentions.

Description

Technical Field

[0001] The present invention relates to artificial intelligence related fields, and specifically to a media content intelligent retrieval system and method based on a large model. Background Art

[0002] In the existing technology, the main technologies used in media content retrieval systems are comprehensive solutions based on keyword queries, knowledge graphs, deep learning, etc., combined with natural language processing. In the field of deep learning, recurrent neural networks (RNNs) are used to process text and sequence data. They use memory units to capture long-term dependencies in sequences to understand complex natural language queries. In addition to deep learning models, natural language processing technology also plays an important role in media content retrieval systems. Text preprocessing technologies such as word segmentation, part-of-speech tagging, and named entity recognition are used to understand the query statements entered by users. Semantic analysis and sentiment analysis technologies are used to further explore the deep meaning behind the query statements, thereby improving the accuracy of retrieval results.

[0003] However, even the relatively mature retrieval technologies mentioned above still have shortcomings that cannot be ignored: traditional natural language processing technologies often rely on manually constructed rules and feature templates, which require a lot of expert knowledge and experience, and have weak adaptability to new application scenarios and languages. They are often unable to cope with complex language phenomena. For example, for complex language phenomena such as long sentences, polysemous words, and metaphors, traditional methods usually need to build complex rules and models, and it is difficult to guarantee the accuracy and efficiency of processing.

[0004] Traditional natural language processing technology also has certain limitations when dealing with multilingual problems. Due to the differences and complexities between different languages, traditional methods require building different models and rules for different languages, which increases the difficulty of development and maintenance.

[0005] The media content retrieval system based on keywords and knowledge graphs relies on keywords and other information extracted by the above-mentioned natural language processing technology. In complex scenarios containing metaphorical information (such as user questions containing emotions, feelings, metaphors, etc.), the information extracted will be irrelevant.

[0006] When a user's multiple questions have contextual relationships, the current keyword and knowledge graph-based technologies find it difficult to understand the user's questions, resulting in retrieval results that do not meet the user's expectations; when the user's questions are irrelevant to the retrieval content implemented by the current system, traditional retrieval technology either cannot answer the user's questions or gives an incorrect answer. Summary of the invention

[0007] The purpose of the present invention is to provide a media content intelligent retrieval system and method based on a large model to solve the problems raised in the above background technology.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] A media content intelligent retrieval system based on a large model, comprising:

[0010] An intention recognition module is used to obtain the search demand data input by the user, process the search demand data, and identify the user intention corresponding to the search demand data through a big data model;

[0011] A data retrieval module is used to convert the user intention obtained through user input based on the big data model, obtain the corresponding search conditions, and search the media database according to the search conditions to obtain the related media data;

[0012] The result generation module is used to use the media data input by the user and the output of the data retrieval module, as well as the historical data as associated data to generate the final result through the large model.

[0013] As a further solution of the present invention: the search requirement data may be text data and audio data, and further includes an intention data normalization module, specifically including:

[0014] A normalization trigger unit, used to judge the retrieval requirement data, and if the retrieval requirement data is audio data, trigger a data normalization program;

[0015] a normalization execution unit, configured to execute a data normalization procedure, identify and match the audio data based on the speech recognition model to convert the audio data into text data, retrieve matching dialect information based on speech features of the audio data, and verify the text data based on the dialect information;

[0016] The emotion normalization unit is used to perform emotion carrying analysis on the audio data, obtain emotion expression information of the audio data, and mark the text data based on the emotion expression information.

[0017] As a further solution of the present invention: the intention recognition module specifically includes:

[0018] A preprocessing unit, configured to receive user input and preprocess the text input by the user based on a large language model, wherein the preprocessing includes removing irrelevant characters, punctuation marks, and text segmentation;

[0019] The intention recognition unit is used to parse the text input by the user, identify the key information in the text, including themes, emotions and opinions, and analyze and judge the keywords, phrases and sentence structures in the text and their logical relationships.

[0020] As a further solution of the present invention: the data retrieval module specifically includes:

[0021] Input parsing unit: When a user enters a query request, the large model performs in-depth analysis on the input, including word segmentation and part-of-speech tagging, and obtains the intent and semantics behind the user's query through natural language processing;

[0022] The query positioning unit is used to generate an SQL query statement that can accurately express the user's intention based on the database structure and table relationship, and send it to the media content database for execution.

[0023] As a further solution of the present invention: the result generation module specifically includes:

[0024] The association and fusion unit is used to associate and fuse the query request input by the user, the media data, and the historical data stored in the system to form a comprehensive and complex context;

[0025] A language output unit is used to process the context based on a generative dialogue technology and generate a result output, wherein the result output is characterized by natural and fluent language data.

[0026] A media content intelligent retrieval method based on a large model, comprising:

[0027] Acquire the search demand data input by the user, process the search demand data, and identify the user intention corresponding to the search demand data through a big data model;

[0028] Based on the big data model, the user intention obtained through user input is converted to obtain corresponding search conditions, and the media database is searched according to the search conditions to obtain related media data;

[0029] The media data input by the user and the output of the data retrieval module, as well as the historical data, are used as associated data to generate the final result through the large model.

[0030] As a further solution of the present invention: the search requirement data may be text data and audio data, and further includes a step of normalizing the intention data, specifically including:

[0031] The search requirement data is judged, and if the search requirement data is audio data, a data normalization procedure is triggered;

[0032] Executing a data normalization procedure, identifying and matching the audio data based on the speech recognition model to convert the audio data into text data, retrieving matching dialect information based on the speech features of the audio data, and verifying the text data based on the dialect information;

[0033] The audio data is subjected to an emotion carrying analysis to obtain the emotion expression information of the audio data, and the text data is marked based on the emotion expression information.

[0034] As a further solution of the present invention: the steps of obtaining the search demand data input by the user, processing the search demand data, and identifying the user intention corresponding to the search demand data through the big data model specifically include:

[0035] Receiving user input, and preprocessing the text input by the user based on the large language model, wherein the preprocessing includes removing irrelevant characters, punctuation marks, and text segmentation;

[0036] Parse the text input by the user, identify the key information in the text, including themes, emotions and opinions, analyze and judge the keywords, phrases and sentence structures in the text and their logical relationships.

[0037] As a further solution of the present invention: the step of converting the user intention obtained through user input based on the big data model, obtaining corresponding search conditions, and searching the media database according to the search conditions to obtain the associated media data specifically includes:

[0038] When a user enters a query request, the large model performs in-depth analysis of the input, including word segmentation and part-of-speech tagging, and obtains the intent and semantics behind the user's query through natural language processing;

[0039] Based on the database structure and table relationships, an SQL query statement that can accurately express the user's intention is generated and sent to the media content database for execution.

[0040] As a further solution of the present invention: the step of using the media data input by the user and the media data output by the data retrieval module, as well as the historical data as the associated data, and generating the final result through the large model specifically includes:

[0041] The query request input by the user, the media data and the historical data stored in the system are associated and integrated to form a comprehensive and complex context;

[0042] The context is processed based on a generative dialogue technique, and a result output is generated, where the result output is represented by natural and fluent language data.

[0043] Compared with the prior art, the beneficial effects of the present invention are: by using a large language model for training based on a large amount of corpus data, it can automatically learn language rules and knowledge without human intervention, so it has stronger adaptability and generalization ability, and through technologies such as multilingual pre-training, it can process multiple languages ​​​​at the same time, and can realize cross-language information transmission and knowledge sharing, realize sentiment analysis of user questions, capture the user's real needs, and identify the user's question intentions, and output corresponding results according to different intentions. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 The module logic diagram of a media content intelligent retrieval system based on a large model.

[0045] Figure 2 This is a structural block diagram of a media content intelligent retrieval system based on a large model.

[0046] Figure 3 The flowchart is a flow chart of a media content intelligent retrieval method based on a large model. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0048] The specific implementation of the present invention is described in detail below in conjunction with specific embodiments.

[0049] like Figure 1 to Figure 2 The above-mentioned media content intelligent retrieval system based on a large model provided by an embodiment of the present invention includes:

[0050] An intention recognition module is used to obtain the search demand data input by the user, process the search demand data, and identify the user intention corresponding to the search demand data through a big data model;

[0051] A data retrieval module is used to convert the user intention obtained through user input based on the big data model, obtain the corresponding search conditions, and search the media database according to the search conditions to obtain the related media data;

[0052] The result generation module is used to use the media data input by the user and the output of the data retrieval module, as well as the historical data as associated data to generate the final result through the large model.

[0053] In this embodiment, a media content intelligent retrieval system based on a large model is provided. By using a large language model for training based on a large amount of corpus data, it can automatically learn language rules and knowledge without human intervention, so it has stronger adaptability and generalization ability, and through multilingual pre-training and other technologies, it can process multiple languages ​​at the same time, and can realize cross-language information transmission and knowledge sharing, realize sentiment analysis of user questions, capture the real needs of users, and identify the user's question intentions, and output corresponding results according to different intentions; Based on the advantages of the solution in this embodiment compared with the existing technology, this technical solution is more suitable for: First, advertising and marketing: helping brands quickly find their positioning and target audiences Related media resources. By analyzing massive media data, the system can identify popular trends and topics, providing strong support for advertising planning. In addition, the system can also be used for precision marketing to push customized advertising content based on user interests and behavioral habits; second, news and media: the system can greatly improve the efficiency of news editing and editing. Editors can quickly retrieve pictures, videos and background information related to news events through the system, providing rich materials for news reports; third, e-commerce: help e-commerce platforms better manage multimedia information such as product pictures, videos and user reviews. By analyzing and organizing this information, the system can provide merchants with suggestions such as product recommendations and pricing strategies to improve sales results.

[0054] As another preferred embodiment of the present invention, the search requirement data may be text data and audio data, and further includes an intention data normalization module, specifically including:

[0055] A normalization trigger unit, used to judge the retrieval requirement data, and if the retrieval requirement data is audio data, trigger a data normalization program;

[0056] a normalization execution unit, configured to execute a data normalization procedure, identify and match the audio data based on the speech recognition model to convert the audio data into text data, retrieve matching dialect information based on speech features of the audio data, and verify the text data based on the dialect information;

[0057] The emotion normalization unit is used to perform emotion carrying analysis on the audio data, obtain emotion expression information of the audio data, and mark the text data based on the emotion expression information.

[0058] Furthermore, the intention recognition module specifically includes:

[0059] A preprocessing unit, configured to receive user input and preprocess the text input by the user based on a large language model, wherein the preprocessing includes removing irrelevant characters, punctuation marks, and text segmentation;

[0060] The intention recognition unit is used to parse the text input by the user, identify the key information in the text, including themes, emotions and opinions, and analyze and judge the keywords, phrases and sentence structures in the text and their logical relationships.

[0061] In this embodiment, the main technical solution is to identify the intention of user input based on a large language model, determine whether the current user input is related to the media content, and output one of the results of yes or no; the implementation method is: set prompt words for the large model, specify the model role tasks, limit its output to a format, and implement the intention recognition of the large model based on the prompt project; the specific process is: after the module receives the user input, it immediately starts its powerful analysis ability based on the large language model itself. First, the text input by the user is preprocessed, including removing irrelevant characters, standardizing punctuation marks, and text segmentation to ensure the accuracy of subsequent analysis; then the text input by the user is parsed. The model is trained through a large amount of text data. It has learned to understand the complexity and diversity of human language and can identify key information in the text, such as themes, emotions, opinions, etc. During the intent recognition stage, the model will focus on the expression of intent in the text; it will analyze the keywords, phrases and sentence structures in the text, as well as the logical relationships between them, to infer the user's true intent; intent recognition based on the large model has the ability to self-learn and optimize. It will continuously adjust and optimize its recognition algorithm and model parameters based on user feedback and system operation data to improve recognition accuracy and efficiency. This capability enables the intent recognition module to continuously adapt to new language phenomena and user behaviors, maintaining its long-term effectiveness and reliability.

[0062] As another preferred embodiment of the present invention, the data retrieval module specifically includes:

[0063] Input parsing unit: When a user enters a query request, the large model performs in-depth analysis on the input, including word segmentation and part-of-speech tagging, and obtains the intent and semantics behind the user's query through natural language processing;

[0064] The query positioning unit is used to generate an SQL query statement that can accurately express the user's intention based on the database structure and table relationship, and send it to the media content database for execution.

[0065] In this embodiment, the main technical scheme is: based on the big model, the user input is converted into search conditions, and the relevant media data is extracted from the media content database through the search conditions; the core of the data retrieval module is to convert the natural language input by the user into precise search conditions, and extract the relevant media data from the huge media content database accordingly; in this process, the big model plays a vital role; when the user enters a query request, the big model will first perform a deep analysis of the input; it is not just a simple word segmentation and part-of-speech tagging of the text, but also understands the intention and semantics behind the user query through natural language processing technology; for example, the user may enter "find news reports about artificial intelligence last year", the big model will identify the keywords "last year", "artificial intelligence" and "news reports", and understand that the user wants to find news about a certain topic within a specific time range; next, the big model generates an SQL query statement that can accurately express the user's intention based on the structure and table relationship of the database, and sends it to the media content database for execution; the database will filter out qualified results from the massive media data based on the conditions in the SQL statement; these results may be media data in various forms such as text, pictures, and videos.

[0066] As another preferred embodiment of the present invention, the result generation module specifically includes:

[0067] The association and fusion unit is used to associate and fuse the query request input by the user, the media data, and the historical data stored in the system to form a comprehensive and complex context;

[0068] A language output unit is used to process the context based on a generative dialogue technology and generate a result output, wherein the result output is characterized by natural and fluent language data.

[0069] In this embodiment, the main technical scheme is: based on the big model prompt project, the media content information input by the user and the output of the data retrieval module, as well as the historical data, are used as context to generate the final result through the big model; in the result generation module, the powerful capabilities of the big model prompt project are fully utilized to integrate the query request input by the user, the media content information accurately screened by the data retrieval module, and the historical data stored in the system to form a comprehensive and complex context; this context construction method ensures that the final generated result is not only accurate, but also can deeply meet the personalized needs of users; in the process of generating results, the big model will adopt the "generative dialogue" technology, which allows the big model to generate natural and fluent language according to the context, so that the final result is not only rich in information, but also easy to understand; at the same time, the generative dialogue technology can also ensure that the big model maintains consistency when generating results, avoiding logical confusion or information omission; in addition, the "diversity enhancement" mechanism is introduced in the result generation module, which allows the big model to consider multiple possible answers or explanations when generating results, and select the best one from them; the advantage of doing so is that it can increase the diversity of results, allowing users to see more interpretations and explanations from different angles.

[0070] like Figure 3 As shown, an embodiment of the present invention provides a method for intelligent retrieval of media content based on a large model, comprising:

[0071] S10, obtaining search demand data input by a user, processing the search demand data, and identifying the user intention corresponding to the search demand data through a big data model;

[0072] S20, converting the user intention obtained through the user input based on the big data model, obtaining corresponding search conditions, and searching the media database according to the search conditions to obtain related media data;

[0073] S30, using the media data input by the user and output by the data retrieval module, as well as the historical data as associated data, and generating a final result through a large model.

[0074] As another preferred embodiment of the present invention, the search requirement data may be text data and audio data, and further includes an intention data normalization step, specifically including:

[0075] The search requirement data is judged, and if the search requirement data is audio data, a data normalization procedure is triggered;

[0076] Executing a data normalization procedure, identifying and matching the audio data based on the speech recognition model to convert the audio data into text data, retrieving matching dialect information based on the speech features of the audio data, and verifying the text data based on the dialect information;

[0077] The audio data is subjected to an emotion carrying analysis to obtain the emotion expression information of the audio data, and the text data is marked based on the emotion expression information.

[0078] As another preferred embodiment of the present invention, the steps of obtaining the search demand data input by the user, processing the search demand data, and identifying the user intention corresponding to the search demand data through the big data model specifically include:

[0079] Receiving user input, and preprocessing the text input by the user based on the large language model, wherein the preprocessing includes removing irrelevant characters, punctuation marks, and text segmentation;

[0080] Parse the text input by the user, identify the key information in the text, including themes, emotions and opinions, analyze and judge the keywords, phrases and sentence structures in the text and their logical relationships.

[0081] As another preferred embodiment of the present invention, the steps of converting the user intention obtained through user input based on the big data model, obtaining corresponding search conditions, and searching the media database according to the search conditions to obtain the associated media data specifically include:

[0082] When a user enters a query request, the large model performs in-depth analysis of the input, including word segmentation and part-of-speech tagging, and obtains the intent and semantics behind the user's query through natural language processing;

[0083] Based on the database structure and table relationships, an SQL query statement that can accurately express the user's intention is generated and sent to the media content database for execution.

[0084] As another preferred embodiment of the present invention, the step of using the media data input by the user and output by the data retrieval module, and the historical data as associated data to generate the final result through the large model specifically includes:

[0085] The query request input by the user, the media data and the historical data stored in the system are associated and integrated to form a comprehensive and complex context;

[0086] The context is processed based on a generative dialogue technique, and a result output is generated, where the result output is represented by natural and fluent language data.

[0087] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0088] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the disclosure in the specification and examples. This application is intended to cover any variations, uses or adaptations of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the claims.

[0089] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A media content intelligent retrieval system based on a large model, characterized in that: Include: An intention recognition module is used to obtain the search demand data input by the user, process the search demand data, and identify the user intention corresponding to the search demand data through a big data model; A data retrieval module is used to convert the user intention obtained through user input based on the big data model, obtain the corresponding search conditions, and search the media database according to the search conditions to obtain the related media data; The result generation module is used to use the media data input by the user and the output of the data retrieval module, as well as the historical data as associated data to generate the final result through the large model.

2. According to claim 1, a media content intelligent retrieval system based on a large model is characterized in that: The search requirement data may be text data and audio data, and also includes an intention data normalization module, specifically including: A normalization trigger unit, used to judge the retrieval requirement data, and if the retrieval requirement data is audio data, trigger a data normalization program; a normalization execution unit, configured to execute a data normalization procedure, identify and match the audio data based on the speech recognition model to convert the audio data into text data, retrieve matching dialect information based on speech features of the audio data, and verify the text data based on the dialect information; The emotion normalization unit is used to perform emotion carrying analysis on the audio data, obtain emotion expression information of the audio data, and mark the text data based on the emotion expression information.

3. A media content intelligent retrieval system based on a large model according to claim 2, characterized in that: The intention recognition module specifically includes: A preprocessing unit, configured to receive user input and preprocess the text input by the user based on a large language model, wherein the preprocessing includes removing irrelevant characters, punctuation marks, and text segmentation; The intention recognition unit is used to parse the text input by the user, identify the key information in the text, including themes, emotions and opinions, and analyze and judge the keywords, phrases and sentence structures in the text and their logical relationships.

4. A media content intelligent retrieval system based on a large model according to claim 3, characterized in that: The data retrieval module specifically includes: Input parsing unit: When a user enters a query request, the large model performs in-depth analysis on the input, including word segmentation and part-of-speech tagging, and obtains the intent and semantics behind the user's query through natural language processing; The query positioning unit is used to generate an SQL query statement that can accurately express the user's intention based on the database structure and table relationship, and send it to the media content database for execution.

5. A media content intelligent retrieval system based on a large model according to claim 4, characterized in that: The result generation module specifically includes: The association and fusion unit is used to associate and fuse the query request input by the user, the media data, and the historical data stored in the system to form a comprehensive and complex context; A language output unit is used to process the context based on a generative dialogue technology and generate a result output, wherein the result output is characterized by natural and fluent language data.

6. A media content intelligent retrieval method based on a large model, characterized in that: Include: Acquire the search demand data input by the user, process the search demand data, and identify the user intention corresponding to the search demand data through a big data model; Based on the big data model, the user intention obtained through user input is converted to obtain corresponding search conditions, and the media database is searched according to the search conditions to obtain related media data; The media data input by the user and the output of the data retrieval module, as well as the historical data, are used as associated data to generate the final result through the large model.

7. A method for intelligent retrieval of media content based on a large model according to claim 6, characterized in that: The search requirement data may be text data and audio data, and also includes a step of normalizing the intent data, specifically including: The search requirement data is judged, and if the search requirement data is audio data, a data normalization procedure is triggered; Execute a data normalization procedure to identify and match the audio data based on the speech recognition model to convert the audio data into text data, retrieve matching dialect information based on the speech features of the audio data, and verify the text data based on the dialect information; Perform emotion carrying analysis on the audio data to obtain emotion expression information of the audio data, and mark the text data based on the emotion expression information.

8. The method for intelligent retrieval of media content based on a large model according to claim 7, characterized in that: The steps of obtaining the search demand data input by the user, processing the search demand data, and identifying the user intention corresponding to the search demand data through the big data model specifically include: Receiving user input, and preprocessing the text input by the user based on the large language model, wherein the preprocessing includes removing irrelevant characters, punctuation marks, and text segmentation; Parse the text input by the user, identify the key information in the text, including themes, emotions and opinions, analyze and judge the keywords, phrases and sentence structures in the text and their logical relationships.

9. The method for intelligent retrieval of media content based on a large model according to claim 8, characterized in that: The steps of converting the user intention obtained through user input based on the big data model, obtaining corresponding search conditions, and searching the media database according to the search conditions to obtain the associated media data specifically include: When a user enters a query request, the large model performs in-depth analysis of the input, including word segmentation and part-of-speech tagging, and obtains the intent and semantics behind the user's query through natural language processing; Based on the database structure and table relationships, an SQL query statement that can accurately express the user's intention is generated and sent to the media content database for execution.

10. The method for intelligent retrieval of media content based on a large model according to claim 9, characterized in that: The step of using the media data input by the user and the media data output by the data retrieval module, as well as the historical data as the associated data, and generating the final result through the large model specifically includes: The query request input by the user, the media data and the historical data stored in the system are associated and integrated to form a comprehensive and complex context; The context is processed based on a generative dialogue technique, and a result output is generated, wherein the result output is characterized by natural and fluent language data.

Images