A water product recommendation method and device based on large model

Abstract

The present invention provides a method and device for recommending water products based on a large model. The method includes: obtaining water industry request information; performing feature extraction on the water industry request information to obtain target water key features; obtaining water feedback information based on the target water key features and a preset water knowledge base; obtaining water product recommendation information based on the target water key features and a preset water product library; and obtaining a feedback report based on the water industry request information, the water feedback information, and the water product recommendation information. The present invention can not only provide professional answers to users in the water industry, but also recommend water industry-related products with a high degree of matching to users based on their intentions, which is conducive to improving user satisfaction and reducing sales costs.

Description

Technical Field

[0001] The present invention relates to the field of water technology, and also to a method and device for recommending water products based on a large model. Background Art

[0002] The water industry encompasses a supply chain encompassing raw water, water supply, water conservation, drainage, sewage treatment, and water resource recycling. Existing question-answering robots or large models have limitations when addressing questions in this specialized field. They require the assistance of professionals within the water industry to effectively understand user intent and provide answers. Furthermore, large models cannot provide products tailored to the user's intent, resulting in a poor user experience and hindering the promotion of related products within the water industry. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a water product recommendation method and device based on a large model, so as to improve the user experience of the water industry and reduce sales costs.

[0004] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0005] A first aspect of the present invention provides a water product recommendation method based on a large model, comprising:

[0006] Obtain information requested by the water industry;

[0007] Extracting features from the water industry request information to obtain key features of the target water industry;

[0008] Obtaining water affairs feedback information based on the target water affairs key characteristics and a preset water affairs knowledge base;

[0009] Obtaining water service product recommendation information based on the target water service key characteristics and the preset water service product library;

[0010] A feedback report is obtained according to the water industry request information, the water feedback information and the water product recommendation information.

[0011] Optionally, obtain water industry request information, including:

[0012] Get the original request information;

[0013] The original request information is filtered according to preset filtering conditions to obtain water industry request information.

[0014] Optionally, feature extraction is performed on the water industry request information to obtain target water industry key features, including:

[0015] Obtain preset water affairs key feature database;

[0016] Performing feature extraction on the water industry request information to obtain original key features;

[0017] Target water affairs key features are determined according to the preset water affairs key feature library and the original key features.

[0018] Optionally, water service feedback information is obtained based on the target water service key characteristics and a preset water service knowledge base, including:

[0019] Access to preset water affairs knowledge base;

[0020] Obtaining water affairs information according to the target water affairs key characteristics and the preset water affairs knowledge base;

[0021] The water affairs information is adjusted to obtain water affairs feedback information.

[0022] Optionally, based on the target water service key features and a preset water service product library, water service product recommendation information is obtained, including:

[0023] Get access to a library of preset water products;

[0024] Obtaining water service product matching information based on the target water service key characteristics and a preset water service product library;

[0025] The water product recommendation information is determined based on the preset recommendation value and the water product matching information.

[0026] Optionally, obtain a library of pre-set water products, including:

[0027] Obtain information on water products on sale;

[0028] Dividing the water utility product information on sale according to preset grouping conditions to obtain divided water utility product information on sale;

[0029] A preset water products library is determined based on the divided water products on sale information.

[0030] Optionally, a feedback report is obtained according to the water industry request information, the water industry feedback information, and the water industry product recommendation information, including:

[0031] Get the feedback report template;

[0032] The feedback report template is filled in according to the water industry request information, the water feedback information and the water product recommendation information to obtain a feedback report.

[0033] A second aspect of the present invention provides a water product recommendation device based on a large model, comprising:

[0034] Acquisition module, used to obtain water industry request information;

[0035] A first processing module is used to extract features from the water industry request information to obtain key features of the target water industry;

[0036] A second processing module is configured to obtain water service feedback information based on the target water service key characteristics and a preset water service knowledge base;

[0037] A third processing module is configured to obtain water service product recommendation information based on the target water service key features and a preset water service product library;

[0038] The fourth processing module is used to obtain a feedback report according to the water industry request information, the water industry feedback information and the water industry product recommendation information.

[0039] According to a third aspect of the present invention, a computing device is provided, comprising: a processor and a memory storing a computer program, wherein when the computer program is executed by the processor, the method according to the first aspect is executed.

[0040] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions, which, when executed on a computer, causes the computer to execute the method described in the first aspect.

[0041] The above solution of the present invention includes at least the following beneficial effects:

[0042] The above-mentioned scheme of the present invention obtains the target water service key features based on the obtained water service industry request information, and then obtains water service feedback information and water service product recommendation information based on the preset water service knowledge base and the preset water service product library. Finally, a feedback report is obtained based on the water service industry request information, water service feedback information and water service product recommendation information. It can not only provide professional answers to users in the water service industry, but also recommend water service industry-related products with a higher matching degree to users based on user intentions, which is conducive to improving user satisfaction and reducing sales costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 1 is a flow chart of a water product recommendation method based on a large model in an embodiment of the present invention;

[0044] Figure 2 4 is a schematic structural diagram of a water product recommendation device based on a large model in an embodiment of the present invention. DETAILED DESCRIPTION

[0045] Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0046] like Figure 1 As shown, an embodiment of the present invention proposes a water product recommendation method based on a large model, comprising the following steps:

[0047] Step 101, obtaining water industry request information;

[0048] Step 102: extracting features from the water industry request information to obtain target water industry key features;

[0049] Step 103: obtaining water service feedback information based on the target water service key features and a preset water service knowledge base;

[0050] Step 104: obtaining water service product recommendation information based on the target water service key features and a preset water service product library;

[0051] Step 105: Obtain a feedback report based on the water industry request information, the water industry feedback information, and the water industry product recommendation information.

[0052] The large-model-based water product recommendation method proposed in an embodiment of the present invention obtains the target water key features based on the acquired water industry request information, and then obtains water feedback information and water product recommendation information based on the preset water knowledge base and the preset water product library. Finally, a feedback report is obtained based on the water industry request information, water feedback information and water product recommendation information. It can not only provide professional answers to users in the water industry, but also recommend water industry-related products with a high degree of matching to users based on user intentions, which is conducive to improving user satisfaction and reducing sales costs.

[0053] In an optional embodiment of the present invention, step 101 includes:

[0054] Step 1011, obtaining original request information;

[0055] Specifically, the original request information is the request information that is input by the user and needs to be answered. The original request information may include content in various formats such as text and numbers. For example, one original request information may be "What is the highest head of a certain water pump?", and another original request information may be "What is the weather like today?". The original request information is subsequently processed.

[0056] Step 1012: Filter the original request information according to preset filtering conditions to obtain water industry request information.

[0057] Specifically, since the method of this embodiment is mainly used to answer request information from the water industry, it is necessary to filter the original request information. The preset filtering condition can be: whether the original request information contains at least one key feature (or keyword) in the preset water key feature library. If so, the original request information is water industry request information, and subsequent steps are processed. For example, the original request information mentioned above is "What is the highest head of a certain water pump?", the "water pump" can be found in the preset water key feature library, so the original request information is water industry request information; if not, for example, the original request information mentioned above is "How is the weather today?", all the key features "today's", "weather", and "how" cannot be found in the preset water key feature library, so the original request information is not water industry request information. "How is the weather today" is input into the existing large model for answering. The large model may feedback information "Today's weather is cloudy", and the output content of the large model "Today's weather is cloudy" is fed back to the user.

[0058] In an optional embodiment of the present invention, step 102 includes:

[0059] Step 1021, obtaining a preset water affairs key feature database;

[0060] Specifically, the preset water affairs key feature library contains a large number of water affairs key features (or keywords) and their corresponding synonyms, misspelled words, pinyin expansion words, etc. Since the water industry request information may contain typos, pinyin, letters, etc., the preset water affairs key feature library not only needs to include a large number of water affairs key features, but also needs to include synonyms, similar fonts, misspelled words, pinyin expansion words, etc. corresponding to the water affairs key features to improve the applicability of the method and user experience.

[0061] Step 1022: extract features from the water industry request information to obtain original key features;

[0062] Specifically, since the water industry request information is filtered information containing key features of the water industry, at least one original key feature can be extracted from the water industry request information. The original key feature may include typos, similar words, and the like.

[0063] Step 1023: Determine target water affairs key features according to the preset water affairs key feature library and the original key features.

[0064] Specifically, the accurate key features corresponding to the original key features can be determined according to the preset water service key feature library as target water service key features for subsequent processing and product recommendation.

[0065] In an optional embodiment of the present invention, step 103 includes:

[0066] Step 1031, obtaining a preset water affairs knowledge base;

[0067] Specifically, the preset water affairs knowledge base contains text materials in various directions within the water affairs industry, such as question banks, instruction manuals, operating manuals, user help, system introductions and other relevant documents and materials in users' daily production activities. The formats are not limited to word, pdf, excel, ppt, photos, web pages, etc. Here, these text materials can also be extracted and classified, such as classifying the above text materials according to different directions such as water plants, sewage plants, secondary water supply, and pipe networks. That is, the preset water affairs knowledge base can contain knowledge bases in different directions such as water plant knowledge base, sewage plant knowledge base, secondary water supply knowledge base, and pipe network knowledge base. After classification, the text materials in each knowledge base are vectorized to achieve the purpose of converting text into vectors and facilitating retrieval and matching.

[0068] Step 1032: Obtain water service information based on the target water service key features and the preset water service knowledge base;

[0069] Specifically, named entity recognition (NER) is performed on knowledge bases in different areas of the pre-set water affairs knowledge base to obtain a pre-set water affairs entity base. Entities with specific meanings are identified within each knowledge base, primarily including names of people, places, organizations, and proper nouns. For example, for a water plant knowledge base, the identified entities may include: equipment names, software system names, business names, contact numbers, contact addresses, equipment parameters, etc. This is primarily used to match the key characteristics of the target water affairs with these entities, and then use these entities to search for corresponding water affairs information in the pre-set water affairs knowledge base.

[0070] Step 1033: Adjust the water service information to obtain water service feedback information.

[0071] Specifically, since the at least one water affairs information obtained may be scattered, it needs to be adjusted to conform to the normal word order. Here, the at least one water affairs information obtained can be input into the existing large model for word order adjustment and arrangement to obtain water affairs feedback information that conforms to the normal word order.

[0072] In an optional embodiment of the present invention, step 104 includes:

[0073] Step 1041, obtaining a preset water product library;

[0074] Specifically, the preset water product library includes all product information related to the water industry on sale, such as product introductions, product manuals, instructions for use, function introductions, web links, etc. Since a large amount of these product information is obtained, they can be classified in different directions, such as dividing the above product information into water plant product library, sewage plant product library, secondary water supply product library and pipe network product library, etc. All these product libraries constitute the preset water product library. Taking the water plant product library as an example, it can include product introductions, product manuals, instructions for use, function introductions, web links and other product information of equipment or software related to the water plant, providing a basis for subsequent product recommendations to users. It should be noted that it is also necessary to establish a corresponding relationship between the preset water product library and the preset water entity library to ensure that the product information in the preset water product library can be obtained through entity searches in the preset water entity library.

[0075] Step 1042: Obtain water service product matching information based on the target water service key features and a preset water service product library;

[0076] Specifically, you can search in the preset water product library based on the target water service key characteristics to obtain corresponding product information. There may be multiple corresponding product information. Here, multiple product information with high correlation with the target water service key characteristics can be output as water service product matching information for subsequent selection.

[0077] Step 1043: Determine water product recommendation information based on the preset recommendation value and the water product matching information.

[0078] Specifically, the water product matching information may contain multiple product information that are highly correlated with the key characteristics of the target water service. Therefore, the product information in the water product matching information can be output according to the preset recommendation value. For example, if the preset recommendation value is 5, the top 5 product information in the water product matching information will be output as water product recommendation information for user selection.

[0079] In an optional embodiment of the present invention, step 1041 includes:

[0080] Step 10411, obtain information on water products on sale;

[0081] Specifically, all information on products on sale in the water industry, such as product introductions, product manuals, instructions for use, function introductions, web links, etc., can be obtained from manufacturers and websites of water industry-related equipment or software.

[0082] Step 10412: Divide the water utility product information on sale according to a preset grouping condition to obtain the divided water utility product information on sale;

[0083] Specifically, the preset grouping conditions are different directions of the water industry. For example, the preset grouping conditions may include: water plant, sewage plant, secondary water supply and pipeline network, etc. The information of water products on sale is divided into the water plant product library and / or sewage plant product library and / or secondary water supply product library and / or pipeline network product library according to the direction to which it belongs. One water product on sale information is classified into at least one product library.

[0084] Step 10413: Determine a preset water product library based on the divided water product information on sale.

[0085] Specifically, after the information on water products on sale is divided, the information on water products on sale in each direction constitutes the product library of this direction, such as the water plant product library, sewage plant product library, secondary water supply product library and pipeline product library, etc. These product libraries constitute the preset water product library.

[0086] In an optional embodiment of the present invention, step 105 includes:

[0087] Step 1051: Obtain a feedback report template;

[0088] Specifically, feedback report templates can be pre-set or obtained as needed to facilitate subsequent filling, quickly generate feedback reports, and improve processing efficiency. Feedback report templates can include user-entered water industry request information, water industry feedback information that answers water industry request information, and water industry product recommendation information based on water industry request information and water industry feedback information.

[0089] Step 1052: Fill in the feedback report template according to the water industry request information, the water feedback information, and the water product recommendation information to obtain a feedback report.

[0090] Specifically, the information required in the feedback report template is extracted from the water industry request information, water feedback information and water product recommendation information, and the feedback report template is filled, so that the feedback report can be generated quickly and effectively.

[0091] A specific embodiment of the water product recommendation method based on a large model proposed in an embodiment of the present invention is as follows:

[0092] The preset water affairs knowledge base in this embodiment may include knowledge bases of different directions in the water affairs industry, such as a water plant knowledge base, a sewage treatment plant knowledge base, a secondary water supply knowledge base, and a pipe network knowledge base, etc.; the preset water affairs entity base may include entity bases of different directions in the water affairs industry, such as a water plant entity base, a sewage treatment plant entity base, a secondary water supply entity base, and a pipe network entity base, etc.; the preset water affairs product base may include product bases of different directions in the water affairs industry, such as a water plant product base, a sewage treatment plant product base, a secondary water supply product base, and a pipe network product base, etc.; for the sake of convenience of explanation, this embodiment takes the water plant knowledge base, the water plant entity base, and the water plant product base as examples for explanation.

[0093] Step 111: Obtain user request information.

[0094] The method of this embodiment is mainly used to process the request information of the water industry. Therefore, after obtaining the user request information, it is necessary to screen the user request information to determine whether it is the request information of the water industry. If so, then perform subsequent processing. Otherwise, input the user request information into the existing question-and-answer robot, and feed back the content output by the question-and-answer robot to the user, which can improve the user experience and the applicability of this method. Specifically, if the user request information contains at least one key feature in the preset water key feature library, then it is determined to be the request information of the water industry.

[0095] Step 112: extract key features of the target water affairs.

[0096] First, it is necessary to determine a preset water service knowledge base. Obtain text materials in different directions within the water service field from channels such as the Internet (considering large scenarios, different directions include water supply plants, sewage treatment plants, secondary water supply, etc.; considering small scenarios, different directions include all equipment in water plants, inspection and maintenance services in water plants, etc.). These text materials include but are not limited to relevant document materials in users' daily production activities such as question banks, user manuals, operation manuals, usage guides, system introductions, etc. The format is not limited to word, pdf, excel, ppt, photos, web pages, etc. Preprocess these text materials, and the preprocessing includes the following steps: 1. Remove stop words: Remove common words such as "of", "is" that are not helpful for classification. 2. Stem extraction / lemmatization: Simplify words to their root forms to reduce lexical diversity. 3. Remove punctuation marks and special characters: Clean non-alphanumeric characters in the text. 4. Lowercase: Convert all text to lowercase to avoid classification errors caused by case differences. 5. Word segmentation: For Chinese text, word segmentation needs to be performed to split sentences into words. After preprocessing, classify these text materials, that is, divide the text materials into at least one knowledge base such as a water supply plant knowledge base, a sewage treatment plant knowledge base, or a secondary water supply knowledge base according to different directions. Then, use an encoder to vectorize the text in each knowledge base for convenient subsequent retrieval and matching. Finally, extract NER entities (named entity recognition such as equipment names, software system names, merchant names, contact numbers, contact addresses, equipment parameters, etc.) for the text content under different knowledge bases.

[0097] After that, determine the corresponding entity library according to the entities identified from each knowledge base, that is, one knowledge base corresponds to one entity library. Taking the water plant knowledge base and the water plant entity library as an example, the entities identified from the water plant knowledge base include: water pipes, valves... etc. These entities such as water pipes, valves... etc. constitute the water plant entity library. Entity libraries in different directions or classifications constitute the preset water service entity library. In addition, it is also necessary to perform synonym, typo, and pinyin expansion on the content of the entity library. For example, the expanded synonyms of secondary water supply include community water supply (business synonym), erci gongshui (typo), secondary gongshui (pinyin expansion). Therefore, the preset water service entity library not only contains the entities identified from the preset water service knowledge base but also includes the synonyms, typos, pinyin expansions, etc. corresponding to each entity.

[0098] In this embodiment, the entity library can be used as a key feature library. That is, the preset water service entity library is also the preset water service key feature library, which is used to perform matching calculations on the original key features extracted from water service industry request information to obtain the target water service key features.

[0099] The water industry request information is vectorized and key features are extracted. The preset water industry key feature library is then calculated based on the original key features. If an entity greater than the preset threshold can be found after the calculation (the calculation can be similar or identical), the entity will be used as the target water industry key feature for subsequent processing.

[0100] Step 113: Determine water service feedback information.

[0101] First, it is necessary to determine a preset water affairs knowledge base. The steps for determining the preset water affairs knowledge base are the same as those for determining the preset water affairs knowledge base in step 112 and will not be repeated here.

[0102] Then, the preset water affairs knowledge base is calculated according to the entity determined in step 112. If at least one text content (i.e., water affairs information) greater than the preset matching value can be found after the calculation (the calculation can be similar or identical), the found water affairs information is input into the existing question-and-answer robot for text sorting, and the sorted text output by the question-and-answer robot can be fed back to the user as water affairs feedback information.

[0103] Step 114: Determine water product recommendation information.

[0104] Product information for different directions in the water industry (i.e., information on water products on sale) is obtained through web pages and manufacturers of water industry products, including but not limited to product introductions, product manuals, instructions for use, function introductions, web links, etc., and the types are not limited to hardware products, software products, value-added services, etc. The product information is divided into the water plant product library and / or sewage plant product library and / or secondary water supply product library and / or pipe network product library according to the direction to which it belongs. One water product on sale information is divided into at least one product library. After the water product on sale information is divided, the water product on sale information in each direction constitutes the product library of this direction, such as the water plant product library, sewage plant product library, secondary water supply product library, pipe network product library, etc. These product libraries constitute the preset water product library.

[0105] Calculation is performed based on the target water service key features and the preset water service product library to obtain corresponding product information. There may be multiple corresponding product information. Multiple product information with high correlation with the target water service key features can be output as water service product matching information, and then the product information in the water service product matching information can be output according to the preset recommendation value. For example, if the preset recommendation value is 5, the top 5 product information in the water service product matching information will be output as water service product recommendation information for user selection.

[0106] To improve the effectiveness of product recommendations, the user's water industry request information can also be input into a trained intent recognition model to determine whether the user intends to purchase related products. If so, water product recommendation information is output. The intent recognition model can be obtained by training a preset network model based on collected user behavior data. The collected user behavior data includes user behavior data on websites or applications, such as browsing history, search keywords, click records, and dwell time. Data with known purchase intentions is extracted from this data as a training set, and the users who ultimately purchased the products and those who did not are labeled as preset labels. The preset network model is trained using the training set and preset labels to obtain a trained intent recognition model. Feature extraction is performed on the user's water industry request information to obtain input features, which are then input into the intent recognition model to obtain a prediction result on whether the user has purchase intention. If the prediction result is that there is purchase intention, water product recommendation information is output to the user.

[0107] Step 115: Determine the feedback report.

[0108] In order to improve the efficiency of generating feedback reports, a feedback report template can be set in advance. The feedback report template may include water industry request information input by the user, water industry feedback information that answers the water industry request information, and water product recommendation information that recommends products based on the water industry request information and water industry feedback information. By extracting the information required in the feedback report template from the water industry request information, water industry feedback information and water product recommendation information, and filling in the feedback report template, the feedback report can be generated quickly and effectively.

[0109] Alternatively, user-entered water industry request information, water industry feedback information that answers water industry request information, and water industry product recommendation information based on water industry request information and water industry feedback information are organized in text form and input into an existing Q&A robot or large model to obtain a summarized feedback report and generate a sharing link. Other users can access the link to quickly understand the conversation conclusion, conversation content, and related product information. The conversation conclusion includes the user-entered water industry request information in the Q&A, water industry feedback information that answers water industry request information, and water industry product recommendation information based on water industry request information and water industry feedback information.

[0110] The specific embodiment of the water product recommendation method based on a large model in the embodiment of the present invention in the direction of water plants in the water industry is as follows:

[0111] Water plant equipment maintenance is crucial to ensuring water quality safety, ensuring water supply stability, improving operational efficiency, and reducing operating costs. Equipment maintenance mainly includes work such as filter replacement, membrane assembly replacement, pump component replacement, and bearing lubrication. Generally, equipment is managed through regular maintenance or daily inspections. During the entire equipment maintenance process, various situations may arise that cannot be resolved by on-site inspectors or engineers. It is necessary to contact the manufacturer or more professional personnel for resolution. During the resolution process, issues such as time cost, communication cost, and personnel cost may arise, leading to low maintenance efficiency and high costs. Taking a water plant as an example, if the water industry request information contains keywords for water hardware products, the processing flow is as follows:

[0112] First, by establishing a knowledge base, entity library, and product library related to water plant equipment, a large-scale question-and-answer approach is used to answer questions about professional knowledge related to equipment maintenance, including maintenance methods, problem resolution, and troubleshooting, to quickly assist in resolving relevant business issues. Furthermore, product recommendations are made for equipment, components, or consumables that appear during multiple rounds of question-and-answer sessions. In this example, textual materials related to water plant equipment are used as knowledge base data, including equipment manuals, repair manuals, maintenance records, maintenance methods, operating instructions, and other related materials.

[0113] Then, in this embodiment, an entity library is established through NER entity recognition, and the entity library contains synonyms corresponding to each entity, etc., where the entity library includes entity id, category name, entity name, entity category, model, purpose and other related information, and synonyms include synonym id, associated category name, entity name, synonym type and other related information.

[0114] Afterwards, in this embodiment, the product information related to the equipment, accessories, consumables, etc. in the water plant field is used as the product library data, including product name, supplier information, product characteristics, accessories list, product parameters and other related information. Establish a relationship between product C in the product library and entity E in the entity library. The degree of correlation is represented by the coefficient β, where b is the offset vector, so that product C can be found in the product library through entity E.

[0115] Afterwards, in this example, when a user asks a question about general knowledge (i.e., not related to the water industry), the existing large model capabilities are directly called to respond, for example: "Introduce Newton"; when a user asks about water equipment-related content, the knowledge base content is matched with the correlation degree, for example: "What is the highest head of a certain water pump?" If there is no relevant information in the knowledge base, the question is directly input into the large model for answering. The knowledge base matching text information is represented by the correlation coefficient α, and the number of text items n retrieved from the knowledge base can be set in advance as needed. Finally, all the obtained text content is input into the existing large model for summary, such as I{i1,i2…i n}=W{α1i1,α2i2…α m i m}|Rank(α) topn Among them, I{i1,i2…i n} is the text set returned by the large model (i.e., the summarized text), n is the number of preset texts; W{α1i1,α2i2…α m i m} are all relevant texts obtained from the knowledge base, the number is m, and they are sorted according to the size of α, α1>α2>α m , make random selection when the coefficients are the same. When m is less than n, the number of returned values ​​is m, Rank(α) topn It means taking the top n texts according to the size of α.

[0116] In the question-answering process, from user questions or large model answers, including large model general ability answers, the degree of relevance between synonyms and entities is represented by the correlation coefficient λ. When the synonym S is identified by the NER algorithm, the final entity E{set} is determined by the coefficient λ as follows:

[0117] E{e1,e2…e j}=S{λ1e1,λ2e2…λ k e k}|Rank(λ) topn

[0118] Among them, E{e1,e2…e j} is the final set of entities, j is the number of entities with the largest correlation coefficients among the first j preset entities S{λ1e1,λ2e2…λ k e k} are all entities corresponding to the identified synonyms, the number is k, and they are sorted according to the size of λ, λ1>λ2>λ k , make random selection when the coefficients are the same. When k is less than j, the number returned is k, Rank(λ) topn It means taking the top j texts according to the size of λ.

[0119] In this embodiment, intent recognition is performed through a trained intent recognition model to determine whether the user has a purchase intention. When the prediction result output by the intent recognition model is that there is no purchase intention, no product recommendation is made. When there is a purchase intention, product information is returned based on the synonym S-entity E-product C association.

[0120] The entities associated with synonyms are:

[0121] E{e1,e2…e j}=S{λ1e1,λ2e2…λ k e k}|Rank(λ) topn

[0122] Append the entities associated with synonyms to the following formula for unified relevance sorting. If it is entity recognition, the associated products are:

[0123] C{c1,c2…c p}=E{β1e1,β2e2…β q e q}|Rank(β) topn

[0124] Among them, C{c1,c2…c p} is the final product set, p is the number of products with the largest correlation coefficients, E{β1e1,β2e2…β q e q} is the entity associated with the identified synonym, the number is q, and it is sorted according to the size of β, β1>β2>β q , make random selection when the coefficients are the same. When q is less than p, the number returned is q, Rank(β) topn Indicates taking the top p entities according to the size of β.

[0125] This embodiment inputs the content of multiple rounds of conversations into a large model to perform text summarization tasks, and finally generates a web link (the link contains a feedback report), which includes product links matched during the conversation process. The person being shared can learn about related content and products through the link.

[0126] The water product recommendation method based on a large model proposed in an embodiment of the present invention performs domain identification on the original information input by the user based on at least one water industry request information of the user, so that the answer result (feedback information) is more accurate; purchase intention is identified through question and answer, and then product recommendation is made, so that product recommendation and target users are more accurate, which is conducive to improving product conversion rate; question and answer based on the water industry knowledge base has more professional information and stronger ability to solve professional problems compared to the existing general large model, while also having the ability to answer general questions; NER recognition based on the synonym library improves the product library hit rate and product library product exposure rate, while improving the input fault tolerance and user experience; using product web page links, target users can obtain more product information, sales communication, order purchase and other subsequent processes more conveniently, reducing the supplier's customer acquisition cost; by sharing summaries, outputting conversation overviews in the form of needs, problems, and solutions can help users make decisions faster and improve product conversion rate.

[0127] like Figure 2 As shown, an embodiment of the present invention proposes a water product recommendation device 200 based on a large model, comprising:

[0128] Acquisition module 201, used to obtain water industry request information;

[0129] The first processing module 202 is used to extract features from the water industry request information to obtain target water industry key features;

[0130] The second processing module 203 is configured to obtain water service feedback information based on the target water service key features and a preset water service knowledge base;

[0131] The third processing module 204 is configured to obtain water service product recommendation information based on the target water service key characteristics and the preset water service product library;

[0132] The fourth processing module 205 is configured to obtain a feedback report based on the water industry request information, the water industry feedback information, and the water industry product recommendation information.

[0133] Optionally, obtain water industry request information, including:

[0134] Get the original request information;

[0135] The original request information is filtered according to preset filtering conditions to obtain water industry request information.

[0136] Optionally, feature extraction is performed on the water industry request information to obtain target water industry key features, including:

[0137] Obtain preset water affairs key feature database;

[0138] Performing feature extraction on the water industry request information to obtain original key features;

[0139] Target water affairs key features are determined according to the preset water affairs key feature library and the original key features.

[0140] Optionally, water service feedback information is obtained based on the target water service key characteristics and a preset water service knowledge base, including:

[0141] Access to preset water affairs knowledge base;

[0142] Obtaining water affairs information according to the target water affairs key characteristics and the preset water affairs knowledge base;

[0143] The water affairs information is adjusted to obtain water affairs feedback information.

[0144] Optionally, based on the target water service key features and a preset water service product library, water service product recommendation information is obtained, including:

[0145] Get access to a library of preset water products;

[0146] Obtaining water service product matching information based on the target water service key characteristics and a preset water service product library;

[0147] The water product recommendation information is determined based on the preset recommendation value and the water product matching information.

[0148] Optionally, obtain a library of pre-set water products, including:

[0149] Obtain information on water products on sale;

[0150] Dividing the water utility product information on sale according to preset grouping conditions to obtain divided water utility product information on sale;

[0151] A preset water products library is determined based on the divided water products on sale information.

[0152] Optionally, a feedback report is obtained according to the water industry request information, the water industry feedback information, and the water industry product recommendation information, including:

[0153] Get the feedback report template;

[0154] The feedback report template is filled in according to the water industry request information, the water feedback information and the water product recommendation information to obtain a feedback report.

[0155] The large-model-based water product recommendation device proposed in an embodiment of the present invention obtains the target water key features based on the acquired water industry request information, and then obtains water feedback information and water product recommendation information based on the preset water knowledge base and the preset water product library. Finally, a feedback report is obtained based on the water industry request information, water feedback information and water product recommendation information. It can not only provide professional answers to users in the water industry, but also recommend water industry-related products with a high degree of matching to users based on user intentions, which is conducive to improving user satisfaction and reducing sales costs.

[0156] It should be noted that the device is a device corresponding to the above method, and all implementations in the above method embodiment are applicable to the embodiment of the device and can achieve the same technical effects, which will not be described in detail in this embodiment.

[0157] An embodiment of the present invention further provides a computing device comprising: a processor and a memory storing a computer program. When the computer program is executed by the processor, the computer program performs the method described in any of the above embodiments. All implementations in the above method embodiments are applicable to the embodiments of this device and can achieve the same technical effects. These are not further described in this embodiment.

[0158] An embodiment of the present invention further provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described in any of the above embodiments. All implementations in the above method embodiments are applicable to the embodiments of the device and can achieve the same technical effects. These are not further described in this embodiment.

[0159] It should be noted that, in the above embodiments, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element. In addition, it should be pointed out that the scope of the methods and devices in the implementation methods of the above embodiments is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in the opposite order according to the functions involved. For example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0160] The above is 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. A water product recommendation method based on a large model, characterized in that: include: Obtain information requested by the water industry; Extracting features from the water industry request information to obtain key features of the target water industry; Obtaining water affairs feedback information based on the target water affairs key characteristics and a preset water affairs knowledge base; Obtaining water service product recommendation information based on the target water service key characteristics and the preset water service product library; Obtaining a feedback report based on the water industry request information, the water industry feedback information, and the water industry product recommendation information; The feature extraction of the water industry request information is performed to obtain the target water industry key features, including: Obtaining a preset water affairs key feature library; the preset water affairs key feature library includes multiple water affairs key features and their corresponding synonyms, misspelled words, and pinyin expanded words; Performing feature extraction on the water industry request information to obtain original key features; Determining target water affairs key features according to the preset water affairs key feature library and the original key features; The water affairs feedback information is obtained based on the target water affairs key features and the preset water affairs knowledge base, including: Obtain a preset water affairs knowledge base; the preset water affairs knowledge base contains text materials in various directions within the water affairs industry, and the text materials include question banks, instruction manuals, operation manuals, user help, and related documents and materials for system introductions, and the formats are not limited to Word, PDF, Excel, PPT, photos, and web pages; extract and classify the text materials, and classify the text materials according to different directions such as water plants, sewage plants, secondary water supply, and pipe networks, that is, the preset water affairs knowledge base contains knowledge bases in different directions such as water plant knowledge base, sewage plant knowledge base, secondary water supply knowledge base, and pipe network knowledge base; perform named entity recognition on the knowledge bases in different directions in the preset water affairs knowledge base to obtain a preset water affairs entity base; one knowledge base corresponds to one entity base; Obtaining water service information based on the target water service key characteristics and the preset water service knowledge base; matching the target water service key characteristics with an entity, and searching for corresponding water service information in the preset water service knowledge base through the entity; Adjusting the water affairs information to obtain water affairs feedback information; According to the target water service key features and the preset water service product library, water service product recommendation information is obtained, including: Obtain a preset water product library; the preset water product library includes information on products related to the water industry on sale, including product introductions, product manuals, instructions for use, function introductions, and web links; classify the product information according to different directions, and divide the product information into a water plant product library, a sewage plant product library, a secondary water supply product library, and a pipe network product library. All these product libraries constitute the preset water product library; through Establish a relationship between product C in the preset water product library and entity E in the preset water entity library, where: is the correlation degree, b is the offset vector, C is the product in the preset water affairs product library, and E is the entity in the preset water affairs entity library; the product information in the preset water affairs product library is obtained by searching the entity in the preset water affairs entity library; Obtaining water service product matching information based on the target water service key characteristics and a preset water service product library; outputting a plurality of product information having a high correlation with the target water service key characteristics as the water service product matching information; Determining water product recommendation information based on preset recommendation values ​​and the water product matching information; outputting product information in the water product matching information as water product recommendation information according to the preset recommendation values; Among them, obtaining the preset water products library includes: Obtain information on water products on sale; Dividing the water utility product information on sale according to preset grouping conditions to obtain divided water utility product information on sale; Determine a preset water service product library based on the divided water service product information on sale; Wherein, a feedback report is obtained according to the water industry request information, the water industry feedback information and the water industry product recommendation information, including: Obtaining a feedback report template; the feedback report template includes water industry request information input by the user, water industry feedback information that answers the water industry request information, and water industry product recommendation information that recommends products based on the water industry request information and the water industry feedback information; Filling the feedback report template according to the water industry request information, the water feedback information and the water product recommendation information to obtain a feedback report; Among them, obtaining water industry request information includes: Obtaining original request information; the original request information is the request information input by the user that needs to be answered, and the original request information includes content in multiple formats; The original request information is filtered according to a preset filtering condition to obtain water industry request information; wherein the preset filtering condition is: whether the original request information contains at least one key feature or keyword in a preset water industry key feature library; if the original request information contains at least one key feature or keyword in the preset water industry key feature library, the original request information is water industry request information; The method further comprises: Intent recognition is performed using a trained intent recognition model to determine whether the user has purchase intent. If the intent recognition model predicts purchase intent, product information is returned based on the synonym S-entity E-product C association. Among them, the entities associated with synonyms are: ； in, is the final set of entities, j is the number of entities with the largest correlation coefficients. All entities corresponding to the identified synonyms, the number is k, sorted by the size of λ, , make random selection when the coefficients are the same; when k is less than j, the number of returned values ​​is k, Indicates based on The size takes the top j texts; Append the entities associated with the synonyms to the following formula for unified relevance sorting. If it is entity recognition, the associated products are: ； in, is the final product set, p is the number of products with the largest correlation coefficients. The number of entities associated with the identified synonyms is q, according to Sort by size, , make random selection when the coefficients are the same; when q is less than p, the number of returned values ​​is q, Indicates based on The size is taken from the top p entities.

2. A water product recommendation device based on a large model, characterized in that: include: Acquisition module, used to obtain water industry request information; A first processing module is used to extract features from the water industry request information to obtain key features of the target water industry; A second processing module is configured to obtain water service feedback information based on the target water service key characteristics and a preset water service knowledge base; A third processing module is configured to obtain water service product recommendation information based on the target water service key features and a preset water service product library; a fourth processing module, configured to obtain a feedback report based on the water industry request information, the water industry feedback information, and the water industry product recommendation information; The feature extraction of the water industry request information is performed to obtain the target water industry key features, including: Obtaining a preset water affairs key feature library; the preset water affairs key feature library includes multiple water affairs key features and their corresponding synonyms, misspelled words, and pinyin expanded words; Performing feature extraction on the water industry request information to obtain original key features; Determining target water affairs key features according to the preset water affairs key feature library and the original key features; The water affairs feedback information is obtained based on the target water affairs key features and the preset water affairs knowledge base, including: Obtain a preset water affairs knowledge base; the preset water affairs knowledge base contains text materials in various directions within the water affairs industry, and the text materials include question banks, instruction manuals, operation manuals, user help, and related documents and materials for system introductions, and the formats are not limited to Word, PDF, Excel, PPT, photos, and web pages; extract and classify the text materials, and classify the text materials according to different directions such as water plants, sewage plants, secondary water supply, and pipe networks, that is, the preset water affairs knowledge base contains knowledge bases in different directions such as water plant knowledge base, sewage plant knowledge base, secondary water supply knowledge base, and pipe network knowledge base; perform named entity recognition on the knowledge bases in different directions in the preset water affairs knowledge base to obtain a preset water affairs entity base; one knowledge base corresponds to one entity base; Obtaining water service information based on the target water service key characteristics and the preset water service knowledge base; matching the target water service key characteristics with an entity, and searching for corresponding water service information in the preset water service knowledge base through the entity; Adjusting the water affairs information to obtain water affairs feedback information; According to the target water service key features and the preset water service product library, water service product recommendation information is obtained, including: Obtain a preset water product library; the preset water product library includes information on products related to the water industry on sale, including product introductions, product manuals, instructions for use, function introductions, and web links; classify the product information according to different directions, and divide the product information into a water plant product library, a sewage plant product library, a secondary water supply product library, and a pipe network product library. All these product libraries constitute the preset water product library; through Establish a relationship between product C in the preset water product library and entity E in the preset water entity library, where: is the correlation degree, b is the offset vector, C is the product in the preset water affairs product library, and E is the entity in the preset water affairs entity library; the product information in the preset water affairs product library is obtained by searching the entity in the preset water affairs entity library; Obtaining water service product matching information based on the target water service key characteristics and a preset water service product library; outputting a plurality of product information having a high correlation with the target water service key characteristics as the water service product matching information; Determining water product recommendation information based on preset recommendation values ​​and the water product matching information; outputting product information in the water product matching information as water product recommendation information according to the preset recommendation values; Among them, obtaining the preset water products library includes: Obtain information on water products on sale; Dividing the water utility product information on sale according to preset grouping conditions to obtain divided water utility product information on sale; Determine a preset water service product library based on the divided water service product information on sale; Wherein, a feedback report is obtained according to the water industry request information, the water industry feedback information and the water industry product recommendation information, including: Obtaining a feedback report template; the feedback report template includes water industry request information input by the user, water industry feedback information that answers the water industry request information, and water industry product recommendation information that recommends products based on the water industry request information and the water industry feedback information; Filling the feedback report template according to the water industry request information, the water feedback information, and the water product recommendation information to obtain a feedback report; Among them, obtaining water industry request information includes: Obtaining original request information; the original request information is the request information input by the user that needs to be answered, and the original request information includes content in multiple formats; The original request information is filtered according to a preset filtering condition to obtain water industry request information; wherein the preset filtering condition is: whether the original request information contains at least one key feature or keyword in a preset water industry key feature library; if the original request information contains at least one key feature or keyword in the preset water industry key feature library, the original request information is water industry request information; Wherein, the third processing module is further used for: Intent recognition is performed using a trained intent recognition model to determine whether the user has purchase intent. If the intent recognition model predicts purchase intent, product information is returned based on the synonym S-entity E-product C association. Among them, the entities associated with synonyms are: ； in, is the final set of entities, j is the number of entities with the largest correlation coefficients. All entities corresponding to the identified synonyms, the number is k, sorted by the size of λ, , make random selection when the coefficients are the same; when k is less than j, the number of returned values ​​is k, Indicates based on The size takes the top j texts; Append the entities associated with the synonyms to the following formula for unified relevance sorting. If it is entity recognition, the associated products are: ； in, is the final product set, p is the number of products with the largest correlation coefficients. The number of entities associated with the identified synonyms is q, according to Sort by size, , make random selection when the coefficients are the same; when q is less than p, the number of returned values ​​is q, Indicates based on The size is taken from the top p entities.

3. A computing device, characterized in that include: A processor and a memory storing a computer program, wherein when the computer program is executed by the processor, the method according to claim 1 is performed.

4. A computer-readable storage medium, characterized in that The invention stores instructions which, when executed on a computer, cause the computer to execute the method according to claim 1 .

Images