Agricultural intelligent live broadcast system and method based on NLP natural language recognition processing

Abstract

The invention discloses an agricultural intelligent live broadcast system and method based on NLP natural language recognition processing, and the system comprises a live broadcast application server which is applied to network live broadcast, and an anchor end and an audience end which are connected with the application server. The live broadcast application server comprises a data processing end and a data access end which are applied to the live broadcast application server, and the data processing end is used for collecting and storing live broadcast voice data of the anchor end and agricultural product growth data collected by a multi-modal data collection end deployed in a planting base; and the live broadcast voice data and the agricultural product growth data are processed by using an NLP natural language and a live broadcast picture is formed, and the audience terminal accesses the data access terminal of the live broadcast application server through a network to obtain the live broadcast picture so as to realize a live broadcast service. According to the invention, the professional live broadcast technology can be quickly identified, and the requirements of clients can be matched according to the question information, so that the live broadcast term identification rate is greatly improved, and the violation rate of live broadcast can be reduced.

Description

Technical Field

[0001] The present invention belongs to the field of digital agriculture and digital live broadcast technology, and specifically relates to an agricultural intelligent live broadcast system and method based on NLP natural language recognition processing. Background Art

[0002] In recent years, with the continuous development of mobile internet, live streaming has become increasingly popular, and live streaming sales have emerged. Live streaming sales have become the most popular sales model, offering a more intuitive sales experience, a real-time interactive sales atmosphere, and efficient sales rates. Live streaming sales can display products to customers in real time, showcasing their styles and functions. While watching shopping live streams, customers can purchase the products being displayed directly within the host's studio. For industrial, light industrial, and non-staple food products, live streaming can better showcase the styles and functions of related products. Agricultural products, however, have long growth cycles and short harvest times. Failure to accelerate harvest and sales during the harvest season can result in missed harvest opportunities or left to rot, causing significant losses to farmers. Consequently, agricultural product sales are scrambling to adapt to the needs of contemporary consumers and market trends by adopting live streaming sales. However, the live broadcast of agricultural products is completely different from the live broadcast of industry, light industry, and non-staple food. The environmental scenes of most agricultural product growers' online live broadcasts are complex and changeable, which leads to limited live broadcast conditions. The live broadcast system is difficult to cover all areas of crop growth and cannot comprehensively describe the agricultural products grown in all planting areas. It is difficult for customers to fully understand the overall picture of crop growth and the status of agricultural products; moreover, the live broadcast professionalism of agricultural product growers is relatively low, and there are language differences and terminology barriers in the description of agricultural products. The live broadcast system's recognition of the agricultural terms described by the growers is not accurate enough, and the characteristics of the described agricultural products are difficult to quickly match customer needs. It is easy to violate regulations in the real-time commentary of the live broadcast, and it is impossible to quickly interact in multiple dimensions on the live broadcast system platform, which seriously restricts the efficiency of the live broadcast. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent agricultural live broadcast system based on NLP natural language recognition and processing. The invention can quickly identify professional live broadcast techniques and match them to customer needs based on question information, greatly improving the recognition rate of live broadcast terminology and reducing the violation rate of live broadcast. To achieve the above purpose, the present invention adopts the following technical effects: According to one aspect of the present invention, the present invention discloses an agricultural intelligent live broadcast system based on NLP natural language recognition processing, the agricultural intelligent live broadcast system includes a live broadcast application server applied to network live broadcast and an anchor end and an audience end connected to the application server, the live broadcast application server includes a data processing end and a data access end applied in the live broadcast application server, the data processing end is used to collect and store the live broadcast voice data of the anchor end and the agricultural product growth data collected by the multimodal data collection end deployed in the planting base, and uses NLP natural language to process the live broadcast voice data and agricultural product growth data and form a live broadcast picture, and the audience end obtains the live broadcast picture through the network access to the data access end of the live broadcast application server 1 to realize the live broadcast service.

[0004] The above scheme is further preferred, wherein the data processing end includes a voice interaction module, an agricultural product data processing module, a live broadcast data association module, a live broadcast data generation module and a live broadcast data feedback unit; the voice interaction module is used to collect live broadcast voice data and parse out standard language text, and then convert the standard language text into dynamic visual marketing content; the agricultural product data processing module is used to classify, process and extract features of agricultural product growth data collected by the multimodal data collection end to obtain key information of agricultural products; the live broadcast data association module associates the standard language text with the key information of agricultural products, and fuses the associated data obtained through processing with the dynamic visual marketing content to obtain format data compatible with multiple live broadcast platforms; the live broadcast data generation module fuses and compresses the format data to generate a live broadcast picture, and pushes the live broadcast picture to the live broadcast platform in real time for live broadcast; the live broadcast data feedback unit tracks user questions and purchase intention information in the live broadcast picture in real time, and pushes product purchase links to the user's audience end through data access.

[0005] The above solution is further preferred, wherein the multimodal data acquisition module includes an environmental sensor component and an image acquisition device for real-time acquisition of agricultural product growth in the planting base.

[0006] The above solution is further preferred, wherein the voice interaction module includes a voice input unit, a semantic parsing unit, a semantic analysis unit and a decision matching unit; the voice input unit converts the voice inputted by the live broadcast into voice graph data; The semantic parsing unit is used to send the speech atlas data into the NLP semantic parsing model for parsing and correction processing to obtain natural language text information; The semantic analysis unit is used to input the natural language text information into the NLP semantic parsing model for semantic analysis, determine the part of speech, meaning of each word and the grammatical relationship between each word, and convert the natural language text information into the corresponding standard language text; The decision matching unit parses the corpus data in the standard language text to obtain live broadcast terms and sales intentions, and matches and integrates the live broadcast terms and sales intentions with pre-sale products to generate dynamic visual marketing content for the pre-sale products.

[0007] The above solution is further preferred, wherein the agricultural product data processing module includes a term identification and classification unit, an environmental data processing unit, a sentiment description unit, a term matching and fusion unit, and a data optimization unit; The term recognition and classification unit is used to recognize standard language texts and obtain characteristic terms of agricultural products; the environmental data processing unit is used to preprocess agricultural product growth data and send the obtained preprocessed data to the AgriBERT engine for training to obtain agricultural knowledge graph data; The sentiment description unit is used to perform semantic analysis on agricultural knowledge graph data, and generate sentiment descriptions of different degrees according to the degree of deviation between environmental data and preset thresholds, and obtain text data of different sentiment descriptions; The term matching fusion unit is used to fuse the sentiment description text data with the agricultural product characteristic terms to form the live broadcast intention classification data in the current live broadcast scenario; The data optimization unit is used to fuse the live broadcast intention classification data with the dynamic visual marketing content to output format data compatible with multiple live broadcast platforms.

[0008] The above scheme is further preferred, wherein the data access includes a sensitive information filtering unit and a sensitive information execution unit, the sensitive information filtering unit is used to detect and analyze sensitive information appearing in the context of the standard language text to obtain specific sensitive terms, and the sensitive information execution unit is used to filter the sensitive terms or replace them with approximate terms, and re-output the safe standard language text after filtering or replacement.

[0009] The above scheme is further preferred, wherein the data processing end also includes a screen switching module, which extracts characteristic terms related to the growth of agricultural products based on user question statements and purchase intention information, and obtains the corresponding screen collected by the multimodal data collection end based on the characteristic terms.

[0010] The above scheme is further preferred, and pushing the product purchase link to the user includes the following steps: dividing the user types according to the historical interaction data, analyzing the user preferences based on the question data of the user types; generating a corresponding product list according to the user preferences and pushing the product list information and purchase link to the user in the live broadcast screen.

[0011] According to another aspect of the present invention, the present invention provides a method for an agricultural intelligent live broadcast system based on NLP natural language recognition processing, the method comprising the following steps: The live broadcast voice data and the collected agricultural product growth data are converted into live broadcast images, and the live broadcast images are pushed to the live broadcast platform in real time for live broadcast; Track user questions and purchase intentions in the live broadcast in real time, and push product purchase links to users.

[0012] The above scheme is further preferred, and the method also includes extracting features of user question statements and purchase intention information, obtaining feature terms associated with the growth of agricultural products, and switching the live broadcast to the agricultural product planting base screen based on the feature terms.

[0013] In summary, the present invention adopts the above technical solution, and the present invention has the following technical effects: This system uses NLP technology to achieve semantic analysis of planting base monitoring data, agricultural knowledge graph-driven content generation and platform compliance optimization, solving technical problems in traditional agricultural live broadcasts such as boring content, lack of professionalism, and high risk of violations. It reduces the problems of low term recognition rate and inability to recognize professional terminology during live broadcasts, greatly improving the recognition rate of professional agricultural live broadcast terms, and the terms recognized in live broadcasts can quickly match customer needs. In addition, the violation rate of live broadcasts is reduced through real-time filtering mechanisms, and relevant questions raised by viewers can be answered in a timely manner, effectively increasing the audience's stay time and realizing the intelligent promotion and deregulation of agricultural live broadcasts. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is the system principle diagram of the agricultural intelligent live broadcast system based on NLP natural language recognition processing of the present invention Figure 2 It is a schematic diagram of the principle of the data processing end of the present invention; Figure 3 Schematic diagram of the voice interaction module of the present invention; Figure 4 It is a schematic diagram of the principle of the agricultural product data processing module of the present invention; In the accompanying figure, there are a live broadcast application server 1, an anchor terminal 2, an audience terminal 3, and a multimodal data collection terminal 4. DETAILED DESCRIPTION

[0015] To make the objectives, technical solutions, and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and by way of preferred embodiments. However, it should be noted that many of the details listed in this specification are merely provided to help the reader gain a thorough understanding of one or more aspects of the present invention, and these aspects of the present invention can be practiced even without these specific details.

[0016] Combine Figure 1 and Figure 2As shown, according to the agricultural intelligent live broadcast system based on NLP natural language recognition of the present invention, the agricultural intelligent live broadcast system includes a live broadcast application server 1 applied to network live broadcast and an anchor terminal 2 and an audience terminal 3 connected to the application server. The live broadcast application server 1 includes a data processing terminal and a data access terminal applied to the live broadcast application server 1. The data processing terminal is used to collect and store the live broadcast voice data of the anchor terminal 2 and the agricultural product growth data collected by the multimodal data collection terminal 4 deployed in the planting base. The data processing terminal uses NLP natural language processing technology to process the live broadcast voice data and agricultural product growth data and form a live broadcast picture. In the present invention, natural language processing (Natural Language Processing , NLP) technology is used for semantic analysis to ensure that various input keywords maintain a high degree of relevance with standard terms. The host terminal 2 and the audience terminal 3 are installed with application components for realizing live interaction. The application components are APP application terminals that can run live broadcasts, Web pages, live broadcast applets, live broadcast plug-ins or components, etc.; the host terminal 2 and the audience terminal 3 are PCs, smart phones, tablet computers, PDAs, laptops, mobile Internet devices MID (Mobile Internet Devices) or wearable devices, etc.; the audience terminal 2 obtains live broadcast images through the data access terminal of the live broadcast application server 1 through the network to realize live broadcast services, and the audience terminal 2 obtains live broadcast data of the host terminal 2 by accessing the data access terminal through the loaded application components. The multimodal data acquisition module includes an environmental sensor component and image acquisition equipment for real-time acquisition of agricultural product growth in a planting base. The environmental sensor component and image acquisition equipment are statically or dynamically set. The planting base is a greenhouse, open farmland, orchard, aquaculture and other scenes. The image acquisition equipment is one or more of a camera, an audio / video collector, a CCD image sensor, a spectral imager or an infrared imager. The environmental sensor component includes a temperature and humidity sensor, a soil sensor, a light intensity sensor, a flue gas sensor, a wind speed sensor, etc. for obtaining various meteorological data during the growth period of agricultural products.

[0017] In the present invention, Figure 2As shown, the data processing end includes a voice interaction module, an agricultural product data processing module, a live broadcast data association module, a live broadcast data generation module and a live broadcast data feedback unit; the voice interaction module is used to collect live broadcast voice data and parse out standard language text, and then convert the standard language text into dynamic visual marketing content, and store the voice data in a semantic database for extraction, analysis and retrieval; the agricultural product data processing module is used to classify, process and extract features of the agricultural product growth data collected by the multimodal data collection end 4 to obtain key information of agricultural products, obtain agricultural products and corresponding agricultural or agricultural product terms to form a knowledge graph of agriculture or agricultural products, that is, key information of agricultural products; agricultural product growth data includes text scripts, images and other materials collected by sensors, these text scripts are semantically parsed to obtain key information, themes, terms, emotions, and the image materials are classified, cropped, identified, quantified and feature extracted to obtain images of each stage of each agricultural product, that is, obtain single-frame images of each stage of growth of each agricultural product, and the single-frame image is a single picture in each type of photo collection or a single-frame picture in a video, that is, the feature image of each stage; The feature image and text script are subjected to text extraction, description generation, text summary and information fusion to generate multimodal description information of the agricultural product growth period that integrates images, text and video, thereby completing the fusion and splicing to form a knowledge graph of agriculture or agricultural products; the live broadcast data association module associates the standard language text and key information of the agricultural product, and fuses the associated data obtained by the processing with dynamic visual marketing content to obtain format data compatible with multiple live broadcast platforms; the live broadcast data generation module fuses and compresses the format data to generate a live broadcast screen, and pushes the live broadcast screen to the live broadcast platform in real time for live broadcast; the live broadcast data feedback unit tracks the user's question statements and purchase intention information in the live broadcast screen in real time, and pushes product purchase links to the user's audience terminal 2 through data access; when pushing product purchase links to users, user types are divided according to historical interaction data, user preferences are analyzed based on the question data of user types, and data information related to the questions are pushed to users according to the user preferences; a corresponding product list is generated according to user preferences, and the product list information and purchase links are pushed to users in the live broadcast screen, so that purchase intentions can be achieved through live broadcast and push of product information.

[0018] In the present invention, Figure 3As shown, the voice interaction module includes a voice input unit, a semantic parsing unit, a semantic analysis unit and a decision matching unit; the voice input unit converts the voice of the live broadcast input into voice graph data, thereby obtaining a live broadcast corpus knowledge base; the semantic parsing unit is used to send the voice graph data into the NLP semantic parsing model for semantic parsing and correction processing to obtain natural language text information; the semantic analysis unit is used to send the natural language text information into the NLP semantic parsing model for semantic analysis, determine the part of speech, meaning of each word and the grammatical relationship between each word, and convert the natural language text information into the corresponding standard language text, perform semantic parsing on the voice graph data based on the NLP natural language, obtain word form, sentence length, word order, etc., and judge the semantic similarity; the decision matching unit parses the corpus data in the standard language text into live broadcast terms and sales intentions, and compares the live broadcast terms, sales intentions with pre-sales Products are matched and integrated to generate dynamic visual marketing content for pre-sale products; in the present invention, the live broadcast corpus knowledge base contains a large amount of voice feature information and product terms, and the voice feature information involves a variety of dialects, regional cultural customs, etc. The NLP semantic parsing model contains various professional terms and multilingual interpretation corpora. By cleaning, correcting, semantic understanding and keyword extraction of the voice graph data, for example, the term recognition processing of "Wogan" (a local citrus variety): thin skin, thick flesh, abundant accumulation, no pesticides, strong and safe, delicious and sweet; corrected to: "Wogan" from a certain place has thin skin, thick flesh and abundant juice, healthy and safe, no pesticides, delicious and sweet; deep analysis: "Wogan" has thin skin, thick flesh and abundant juice, and the Vc content of Wogan juice is 24 mg / 100 ml; the voice graph data is cleaned to eliminate environmental noise, separate human voices from environmental noise, correct ambiguous words in dialects, and complete real-time conversion to form standard live broadcast terms and sales intentions.

[0019] In the present invention, Figure 4As shown, the agricultural product data processing module includes a term identification and classification unit, an environmental data processing unit, an emotional description unit, a term matching and fusion unit, and a data optimization unit; the term identification and classification unit is used to perform three-level matching and recognition (core library, extended library, fuzzy matching library) on the standard language text to obtain characteristic terms of agricultural products; the information mentioned in the present invention includes the following: core library [Wogan (related varieties)], extended library [Wogan (origin), organic cultivation], fuzzy matching library [limited time offer: 66 yuan / box (10 jin) for the first 100 orders, quality assurance: one bad fruit will be compensated with three]; the environmental data processing unit is used to preprocess the agricultural product growth data and send the obtained preprocessed data to AgriBER The AgriBERT engine is trained to obtain agricultural knowledge graph data; the entities, relationships, and entity attributes of agricultural product growth data are displayed in the form of a graph, and the Neo4j database is used to store agricultural product growth data in the form of a network; the AgriBERT engine has a multi-task output head and a 24-layer Transformer encoder. The 24-layer Transformer encoder trains the preprocessed data and outputs the trained data through the multi-task output head. Multi-task agricultural knowledge graph data with entity recognition, relationship extraction, sentiment analysis, and intent recognition. The agricultural knowledge graph data contains agricultural general word vectors and agricultural professional word vectors, such as pest and disease entity recognition, soil condition recognition, planting technology, agricultural products Professional terms such as yield and quality, agricultural product price trend forecast, and agricultural product policy sentiment analysis; the sentiment description unit is used to perform semantic analysis on agricultural knowledge graph data, and generate sentiment descriptions of different degrees according to the degree of deviation between environmental data and preset thresholds, and obtain text data of different sentiment descriptions; when the degree of deviation between the detection value of soil pH value in the planting environment and the preset threshold; when the real-time environmental parameters (temperature, humidity, soil pH value, light intensity, etc.) deviate from the preset threshold parameters, the current parameter deviation value is calculated: deviation degree = |real-time value-preset threshold median| / (threshold upper limit-threshold lower limit); if the deviation degree is 0-5%, the sentiment level is peace of mind, and the sentiment description keywords are: ideal, stable, healthy; if the deviation An emotional level of 5%-10% is considered "concern," and the key words for describing the emotion are: slight fluctuations, requiring continuous observation. An emotional level of 10-20% is considered "worry," and the key words for describing the emotion are: significant deviation, potential risk, requiring immediate intervention. The term matching and fusion unit is used to fuse the emotional description text data with the characteristic terms of agricultural products to form the live broadcast intention classification data for the current live broadcast scenario. The emotional description text data and the characteristic terms of agricultural products are fused and then implanted into the key information of agricultural products to form the key core live broadcast content. During the live broadcast process, associations are formed between the live broadcast content and question statements, which can better extract the question keywords. The question keywords are used to query and retrieve the content to be broadcast and switch the screen.The data optimization unit is used to fuse the live broadcast intention classification data with dynamic visual marketing content to output format data compatible with multiple live broadcast platforms. The formed format data can be implanted into various live broadcast platforms for simultaneous live broadcast. The generated data script is pushed to the live broadcast stream through regular script compliance processing to complete the live broadcast on each live broadcast platform, thereby reducing the violation rate. By generating different emotional description text data from agricultural knowledge graph data and fusing it with the corresponding agricultural product feature data, the live broadcast intention classification data for the current live broadcast scene is formed. It can ensure that the agricultural product planting base picture can be formed during the screen switching process during the preparation process, and various different emotional description data during the planting process can be obtained, so as to more clearly understand the relevant conditions and data parameters of the agricultural product planting process.

[0020] like Figure 1 As shown, the data access includes a sensitive information filtering unit and a sensitive information execution unit. The sensitive information filtering unit is used to detect and analyze sensitive information that appears in the context of the standard language text, obtain specific sensitive terms, compare the similarity of the sensitive terms with a preset similarity, and compare the sensitivity with a preset sensitivity; if the similarity is greater than the preset similarity and the sensitivity is greater than the preset sensitivity, semantic replacement is performed; the sensitive information execution unit is used to filter the sensitive terms or replace them with similar terms, and re-output the standard language text after filtering or replacement. The sensitive terms are direct sensitive terms, indirect sensitive terms (when the semantic similarity is greater than 0.8), and potential sensitive terms. Direct sensitive terms can be directly filtered and blocked (for example, user account: 138****0123). Indirect sensitive terms and potential sensitive terms are semantically replaced when the term similarity is greater than 0.8 and the sensitivity is greater than 0.7. During the live broadcast of agricultural product planting, growth, or harvesting, sensitive words that appear, such as absolute purification terms (100% freshness preservation) and the use of (pesticide formula) to kill pests, are filtered or replaced with organic additives that can preserve freshness or effectively control pests.

[0021] In the present invention, Figure 2 As shown, the data processing end also includes a screen switching module. The screen switching module extracts characteristic terms associated with agricultural product growth based on user questions and purchase intention information, obtains the corresponding screens collected by the multimodal data acquisition end 4 based on the characteristic terms, and switches the live broadcast screen displayed on the viewer terminal to the agricultural product planting base screen corresponding to the characteristic terms associated with agricultural product growth. During the live broadcast, the agricultural product planting base screen is switched according to the question statement, thereby enabling the most intuitive understanding of the agricultural information of the agricultural products. Through this intuitive agricultural information, it is possible to see the growth change images, growth data, farmland quality, crop varieties, cultivation techniques, agricultural product quality, etc. of the agricultural products during the planting process.

[0022] According to another aspect of the present invention, Figures 1 to 4 The present invention provides a live broadcast method of an agricultural intelligent live broadcast system based on NLP natural language recognition, comprising the following steps: obtaining live broadcast voice data and collected agricultural product growth data to form a live broadcast screen, and streaming the live broadcast screen to a live broadcast platform in real time for live broadcast; tracking user question statements and purchase intention information on the live broadcast screen in real time, and performing feature extraction based on the user question statements and purchase intention information to obtain feature terms associated with agricultural product growth, switching the live broadcast to an agricultural product planting base screen based on the feature terms, and pushing a product purchase link to the user in the live broadcast screen. Based on the characteristics of "Wogan" in the user's question (such as "What is the quality of "Wogan", I want 3 boxes of "Wogan") + historical click behavior, and matching the corresponding products and pushing the stream in real time, according to the "Wogan" characteristics in the question, switch the live screen, and display the planting screen, quality characteristics, and price comparison table of the planting base "Wogan" related live screen (keep the current screen for at least 20s), then switch the screen, and push the purchase link to the user; convert the question response time into purchase conversion rate, and continuously ask questions to switch the planting screen in real time to enhance trust. The present invention realizes the closed loop of data collection → AI decision-making → live screen content generation → commercial conversion, and reshapes the core competitiveness of agricultural product live broadcast by transforming unprofessional live broadcast content into professional knowledge and deep integration with real-time interactive technology.

[0023] If the integrated unit is implemented in the form of a software program module and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a memory and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method. A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the relevant hardware through a program, and the program can be stored in a computer-readable memory. The memory involved in the present invention includes various media that can store program codes, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0024] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. The agricultural intelligent live broadcast system based on NLP natural language recognition processing is characterized by: The agricultural intelligent live broadcast system includes a live broadcast application server applied to network live broadcast and an anchor terminal and an audience terminal connected to the application server. The live broadcast application server includes a data processing terminal and a data access terminal applied in the live broadcast application server. The data processing terminal is used to collect and store the live broadcast voice data of the anchor terminal and the agricultural product growth data collected by the multimodal data collection terminal deployed in the planting base, and uses NLP natural language to process the live broadcast voice data and agricultural product growth data to form a live broadcast picture. The audience terminal obtains the live broadcast picture through the network access to the data access terminal of the live broadcast application server to realize the live broadcast service.

2. The agricultural intelligent live broadcast system based on NLP natural language recognition processing according to claim 1 is characterized in that: The data processing end includes a voice interaction module, an agricultural product data processing module, a live broadcast data association module, a live broadcast data generation module and a live broadcast data feedback unit; the voice interaction module is used to collect live broadcast voice data and parse out standard language text, and then convert the standard language text into dynamic visual marketing content; the agricultural product data processing module is used to classify, process and extract features of agricultural product growth data collected by the multimodal data collection end to obtain key information of agricultural products; the live broadcast data association module associates the standard language text with the key information of agricultural products, and fuses the associated data obtained through processing with the dynamic visual marketing content to obtain format data compatible with multiple live broadcast platforms; the live broadcast data generation module fuses and compresses the format data to generate a live broadcast picture, and pushes the live broadcast picture to the live broadcast platform in real time for live broadcast; The live broadcast data feedback unit tracks user question statements and purchase intention information on the live broadcast screen in real time, and pushes product purchase links to the user's audience terminal through data access.

3. The agricultural intelligent live broadcast system based on NLP natural language recognition processing according to claim 1 is characterized in that: The multimodal data acquisition module includes an environmental sensor component and an image acquisition device for real-time acquisition of the growth of agricultural products in the planting base.

4. The agricultural intelligent live broadcast system based on NLP natural language recognition processing according to claim 2 is characterized in that: The voice interaction module includes a voice input unit, a semantic parsing unit, a semantic analysis unit and a decision matching unit; the voice input unit converts the voice input of the live broadcast into voice graph data; the semantic parsing unit is used to send the voice graph data into the NLP semantic parsing model for parsing and correction processing to obtain natural language text information; the semantic analysis unit is used to send the natural language text information into the NLP semantic parsing model for semantic analysis, determine the part of speech, meaning of each word and the grammatical relationship between each word, and convert the natural language text information into the corresponding standard language text; the decision matching unit parses the corpus data in the standard language text to obtain live broadcast terms and sales intentions, and matches and integrates the live broadcast terms and sales intentions with pre-sale products to generate dynamic visual marketing content for pre-sale products.

5. The agricultural intelligent live broadcast system based on NLP natural language recognition processing according to claim 3 is characterized by: The agricultural product data processing module includes a term identification and classification unit, an environmental data processing unit, a sentiment description unit, a term matching and fusion unit, and a data optimization unit; The term recognition and classification unit is used to recognize standard language text and obtain characteristic terms of agricultural products; the environmental data processing unit is used to preprocess agricultural product growth data, and send the obtained preprocessed data to the AgriBERT engine for training to obtain agricultural knowledge graph data; the emotion description unit is used to perform semantic analysis on agricultural knowledge graph data, and generate emotion descriptions of different degrees according to the degree of deviation between environmental data and a preset threshold, and obtain different emotion description text data; the term matching and fusion unit is used to fuse the emotion description text data with the characteristic terms of agricultural products to form live broadcast intention classification data in the current live broadcast scenario; the data optimization unit is used to fuse the live broadcast intention classification data with dynamic visual marketing content to output format data compatible with multiple live broadcast platforms.

6. The agricultural intelligent live broadcast system based on NLP natural language recognition processing according to claim 2 is characterized in that: The data access includes a sensitive information filtering unit and a sensitive information execution unit. The sensitive information filtering unit is used to detect and analyze sensitive information appearing in the context of a standard language text to obtain specific sensitive terms. The sensitive information execution unit is used to filter sensitive terms or replace them with approximate terms, and re-output the safe standard language text after filtering or replacement.

7. The agricultural intelligent live broadcast system based on NLP natural language recognition processing according to claim 1 or 4, characterized in that: The data processing end also includes a screen switching module, which extracts characteristic terms related to the growth of agricultural products based on user question statements and purchase intention information, and obtains the corresponding screens collected by the multimodal data collection end based on the characteristic terms.

8. The agricultural intelligent live broadcast system based on NLP natural language recognition processing according to claim 2 is characterized in that: Pushing a product purchase chain to users includes the following steps: Categorize user types based on historical interaction data, and analyze user preferences based on question data from user types; Generate a corresponding product list based on user preferences and push the product list information and purchase link to the user in the live broadcast screen.

9. A method for an agricultural intelligent live broadcasting system based on NLP natural language recognition processing suitable for any one of claims 1 to 8, characterized in that: The steps include: The live broadcast voice data and the collected agricultural product growth data are converted into live broadcast images, and the live broadcast images are pushed to the live broadcast platform in real time for live broadcast; Track user questions and purchase intentions in the live broadcast in real time, and push product purchase links to users.

10. The method according to claim 9, characterized in that: The method also includes extracting features from user question statements and purchase intention information, obtaining feature terms associated with the growth of agricultural products, and switching the live broadcast to a screen of the agricultural product planting base based on the feature terms.

Images