Meteorological service key point information prompting method and system

Abstract

The invention relates to the field of information retrieval, in particular to a meteorological service key point information prompting method and system. The method comprises the following steps: acquiring a natural language description of a personalized meteorological information service demand input by a user, extracting basic information of the user, performing semantic embedding and user group identification on the basic information of the user by using a text information processing and coding technology based on artificial intelligence and a large language model, and then performing user group identification on the basic information of the user; user group feature description is carried out on user group identification obtained through identification, and semantic coding is carried out on a user group feature description text generated after group description and natural language description of personalized weather information service requirements; therefore, recommendation content prompted by meteorological service key point information is intelligently generated according to bidirectional semantic attention interaction representation between the semantic coding features described by the user group features and the semantic coding features required by the personalized meteorological information service.

Description

Technical Field

[0001] This application relates to the field of information retrieval, and particularly to a method and system for prompting key points of meteorological service information. Background Art

[0002] In recent years, with the development of the economic society and the improvement of people's living standards, the demand for meteorological information in all walks of life has become increasingly refined and diversified. There are significant differences in the focus and usage methods of weather forecasts among different industries (such as agriculture, transportation, energy, etc.) and user groups (such as farmers, outdoor sports enthusiasts, urban residents, etc.). Therefore, providing personalized and accurate meteorological information services has become the key to enhancing the user experience and service quality.

[0003] However, grass-roots departments, as the main body directly providing meteorological services to users, are facing the dilemma of lacking personnel and technical resources. In terms of personnel, there is a lack of compound talents who are proficient in both meteorological professional knowledge and information technology, making it difficult for grass-roots departments to deeply understand user needs when communicating with users. For example, when facing users engaged in specialty agricultural planting, due to the lack of professional understanding of the complex relationship between meteorology and agricultural production, the staff cannot accurately grasp the deep-level needs of users regarding how meteorological conditions affect crop growth, yield, and quality. In terms of technical resources, there is a lack of advanced data analysis tools, efficient algorithms, and powerful computing capabilities, resulting in the inability to process a large amount of user demand information in a timely and accurate manner, seriously affecting the provision of personalized services.

[0004] Therefore, an optimized scheme for prompting key points of meteorological service information is expected. Summary of the Invention

[0005] This application is made in consideration of the above problems. An object of this application is to provide a method and system for prompting key points of meteorological service information.

[0006] Embodiments of this application provide a method for prompting key points of meteorological service information, which includes:

[0007] Obtaining a natural language description of the personalized meteorological information service requirements input by the user;

[0008] Extracting the basic information of the user, and based on the basic information of the user, performing user group identification on the user to obtain a user group identification result;

[0009] Performing user feature description on the user group identification result to obtain a user group feature description text;

[0010] Semantically encode the user group characteristic description text and the natural language description of the personalized meteorological information service requirements respectively to obtain the user group characteristic description semantic encoding features and the personalized meteorological information service requirements semantic encoding features;

[0011] Perform bidirectional semantic association attention interaction on the user group characteristic description semantic encoding features and the personalized meteorological information service requirements semantic encoding features to obtain user group characteristic-personalized requirement semantic interaction encoding features, including: performing a homography projection transformation on the user group characteristic description semantic encoding features and the personalized meteorological information service requirements semantic encoding features to obtain user group characteristic description homography projection encoding features and personalized meteorological information service requirements homography projection encoding features; performing bidirectional attention field modulation response on the user group characteristic description homography projection encoding features and the personalized meteorological information service requirements homography projection encoding features to obtain the user group characteristic-personalized requirement semantic interaction encoding features;

[0012] Based on the user group characteristic-personalized requirement semantic interaction encoding features, obtain the recommended content of the meteorological service key point information prompt.

[0013] For example, in the meteorological service key point information prompt method according to the embodiment of the present application, wherein, performing user group identification on the user based on the user's basic information to obtain a user group identification result, including:

[0014] Perform semantic embedding encoding on the user's basic information to obtain a user basic information semantic embedding encoding feature vector;

[0015] Input the user basic information semantic embedding encoding feature vector into a user group identifier based on a classifier to obtain the user group identification result.

[0016] For example, in the meteorological service key point information prompt method according to the embodiment of the present application, wherein, performing user characteristic description on the user group identification result to obtain a user group characteristic description text, including: inputting the user group identification result into a user group characteristic descriptor based on a large language model to obtain the user group characteristic description text.

[0017] For example, in the method for prompting meteorological service key point information according to an embodiment of the present application, semantic encoding is respectively performed on the user group characteristic description text and the natural language description of the personalized meteorological information service requirement to obtain a user group characteristic description semantic encoding feature and a personalized meteorological information service requirement semantic encoding feature, including: using a semantic encoder based on Bert-BiLSTM to respectively perform semantic encoding on the user group characteristic description text and the natural language description of the personalized meteorological information service requirement to obtain a user group characteristic description semantic encoding vector as the user group characteristic description semantic encoding feature and a personalized meteorological information service requirement semantic encoding vector as the personalized meteorological information service requirement semantic encoding feature.

[0018] For example, in the method for prompting meteorological service key point information according to an embodiment of the present application, a homography projection transformation is performed on the user group characteristic description semantic encoding feature and the personalized meteorological information service requirement semantic encoding feature to obtain a user group characteristic description homography projection encoding feature and a personalized meteorological information service requirement homography projection encoding feature, including:

[0019] Performing a homography projection transformation on the user group characteristic description semantic encoding vector using a user group characteristic description mapping homography matrix to obtain a user group characteristic description homography projection encoding feature vector as the user group characteristic description homography projection encoding feature;

[0020] Performing a homography projection transformation on the personalized meteorological information service requirement semantic encoding vector using a personalized meteorological information service requirement mapping homography matrix to obtain a personalized meteorological information service requirement homography projection encoding feature vector as the personalized meteorological information service requirement homography projection encoding feature.

[0021] For example, in the method for prompting meteorological service key point information according to an embodiment of the present application, a bidirectional attention field modulation response is performed on the user group characteristic description homography projection encoding feature and the personalized meteorological information service requirement homography projection encoding feature to obtain the user group characteristic-personalized requirement semantic interaction encoding feature, including:

[0022] Constructing a positive and negative user group characteristic-personalized requirement bidirectional attention balance field between the user group characteristic description homography projection encoding feature vector and the personalized meteorological information service requirement homography projection encoding feature vector;

[0023] Map the user group feature description homographic projection encoded feature vector and the personalized meteorological information service demand homographic projection encoded feature vector to the positive and negative user group feature-personalized demand bidirectional attention balance field respectively to obtain the user group feature description homographic projection attention semantic modulation encoded vector and the personalized meteorological information service demand homographic projection attention semantic modulation encoded vector;

[0024] Calculate the element-wise division between the user group feature description homographic projection attention semantic modulation encoded vector and the personalized meteorological information service demand homographic projection attention semantic modulation encoded vector to obtain the user group feature-personalized demand semantic interaction encoded vector as the user group feature-personalized demand semantic interaction encoded feature.

[0025] For example, in the meteorological service key point information prompting method according to the embodiment of the present application, wherein constructing the positive and negative user group feature-personalized demand bidirectional attention balance field between the user group feature description homographic projection encoded feature vector and the personalized meteorological information service demand homographic projection encoded feature vector includes:

[0026] Calculate the forward user group feature-personalized demand attention score field of the user group feature description homographic projection encoded feature vector relative to the personalized meteorological information service demand homographic projection encoded feature vector;

[0027] Calculate the reverse user group feature-personalized demand attention score field of the personalized meteorological information service demand homographic projection encoded feature vector relative to the user group feature description homographic projection encoded feature vector;

[0028] Perform feature splicing and convolutional kernel encoding on the forward user group feature-personalized demand attention score field and the reverse user group feature-personalized demand attention score field to obtain the positive and negative user group feature-personalized demand bidirectional attention balance field.

[0029] For example, in the meteorological service key point information prompting method according to the embodiment of the present application, wherein based on the user group feature-personalized demand semantic interaction encoded feature, obtaining the recommended content for meteorological service key point information prompting includes: after adding a prompt word to the tail of the user group feature-personalized demand semantic interaction encoded vector, inputting it into the meteorological service key point information generator based on the large language model to obtain the recommended content for meteorological service key point information prompting.

[0030] The embodiment of the present application also provides a meteorological service key point information prompting system, which includes:

[0031] A requirement acquisition module for obtaining a natural language description of personalized meteorological information service requirements input by a user;

[0032] A user group identification module for extracting the basic information of the user and identifying the user group based on the basic information of the user to obtain a user group identification result;

[0033] A user feature description module for describing the user group identification result to obtain a user group feature description text;

[0034] A semantic encoding module for respectively performing semantic encoding on the user group feature description text and the natural language description of the personalized meteorological information service requirements to obtain a user group feature description semantic encoding feature and a personalized meteorological information service requirements semantic encoding feature;

[0035] A bidirectional semantic association attention interaction module for performing bidirectional semantic association attention interaction on the user group feature description semantic encoding feature and the personalized meteorological information service requirements semantic encoding feature to obtain a user group feature-personalized requirement semantic interaction encoding feature; wherein, the bidirectional semantic association attention interaction module includes: a homography projection transformation unit for performing homography projection transformation on the user group feature description semantic encoding feature and the personalized meteorological information service requirements semantic encoding feature to obtain a user group feature description homography projection encoding feature and a personalized meteorological information service requirements homography projection encoding feature; a bidirectional attention field modulation response unit for performing bidirectional attention field modulation response on the user group feature description homography projection encoding feature and the personalized meteorological information service requirements homography projection encoding feature to obtain the user group feature-personalized requirement semantic interaction encoding feature;

[0036] A recommendation module for obtaining recommended content for meteorological service key point information prompts based on the user group feature-personalized requirement semantic interaction encoding feature.

[0037] For example, a meteorological service key point information prompt system according to an embodiment of the present application, wherein the user group identification module includes:

[0038] A semantic embedding encoding unit for performing semantic embedding encoding on the basic information of the user to obtain a user basic information semantic embedding encoding feature vector;

[0039] A user group identification unit for inputting the user basic information semantic embedding encoding feature vector into a user group identifier based on a classifier to obtain the user group identification result.

[0040] A method and system for prompting meteorological service key point information according to an embodiment of the present application can automatically process natural language descriptions input by users, not only reducing the demand for compound talents in grass-roots departments, but also being able to efficiently and accurately understand user needs to ensure that the output content meets the specific needs of individual user groups. Moreover, by using advanced data analysis tools and efficient algorithms, it can quickly respond to user needs and provide timely and accurate service suggestions. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the accompanying drawings of the embodiments of the present application will be briefly introduced below. Obviously, the accompanying drawings in the following description only relate to some embodiments of the present application and do not limit the present application.

[0042] Figure 1 Shows a flowchart of the method for prompting meteorological service key point information in an embodiment of the present application;

[0043] Figure 2 Shows a flowchart of sub-step S150 of the method for prompting meteorological service key point information in an embodiment of the present application;

[0044] Figure 3 Shows a flowchart of sub-step S152 of the method for prompting meteorological service key point information in an embodiment of the present application; and

[0045] Figure 4 Shows a schematic structural diagram of the system for prompting meteorological service key point information in an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0046] The terms used in this specification are those general terms that are currently widely used in the art in consideration of the functions of the present application, but these terms may change according to the intentions of those of ordinary skill in the art, precedents, or new technologies in the art. In addition, specific terms may be selected, and in such cases, their detailed meanings will be described in the detailed description of the present application. Therefore, the terms used in the specification should not be understood as simple names, but based on the meanings of the terms and the overall description of the present application.

[0047] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on user terminals and / or servers. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0048] In this application, flowcharts are used to illustrate the operations performed by the system according to the embodiments of this application. It should be understood that the operations before or below do not necessarily need to be executed precisely in sequence. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or several operations can be removed from these processes.

[0049] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only partial embodiments of this application, rather than all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts also belong to the scope of protection of this application.

[0050] In response to the above technical problems, the technical concept of this application is to obtain a natural language description of the personalized meteorological information service requirements input by the user, extract the basic information of the user, use text information processing and encoding technologies based on artificial intelligence and large language models to perform semantic embedding and user group identification on the basic information of the user, then describe the user group characteristics of the identified user group, and then perform semantic encoding on the user group characteristic description text generated after group description and the natural language description of the personalized meteorological information service requirements. In this way, according to the bidirectional semantic attention interaction representation between the semantic encoding features of the user group characteristics description and the semantic encoding features of the personalized meteorological information service requirements, recommended content for intelligent generation of meteorological service key information prompts is generated. This application can automatically process the natural language description input by the user, not only reducing the demand for compound talents in grass-roots departments, but also being able to efficiently and accurately understand the user's needs, ensuring that the output content meets the specific needs of individual user groups. Moreover, by using advanced data analysis tools and efficient algorithms, it can quickly respond to user needs and provide timely and accurate service suggestions.

[0051] Figure 1 A flowchart of the method for prompting meteorological service key information in the embodiments of this application is shown. As Figure 1As shown, the method for prompting key points of meteorological service information according to an embodiment of the present application includes the steps of: S110, obtaining a natural language description of the personalized meteorological information service requirements input by the user; S120, extracting the basic information of the user, and performing user group identification on the user based on the basic information of the user to obtain a user group identification result; S130, performing user feature description on the user group identification result to obtain a user group feature description text; S140, respectively performing semantic encoding on the user group feature description text and the natural language description of the personalized meteorological information service requirements to obtain a user group feature description semantic encoding feature and a personalized meteorological information service requirements semantic encoding feature; S150, performing two-way semantic association attention interaction on the user group feature description semantic encoding feature and the personalized meteorological information service requirements semantic encoding feature to obtain a user group feature-personalized demand semantic interaction encoding feature; S160, based on the user group feature-personalized demand semantic interaction encoding feature, obtaining the recommended content for prompting key points of meteorological service information.

[0052] Specifically, in the technical solution of the present application, first, a natural language description of the personalized meteorological information service requirements input by the user is obtained.

[0053] Next, the basic information of the user is extracted. In the meteorological service system, extracting the basic information of the user is to better understand the user's needs and provide personalized services.

[0054] After that, considering that the basic information of the user includes various types of data, such as age, occupation, interests, geographical location, historical meteorological service usage records, etc. There is rich semantic information between these information. Therefore, in order to capture and refine the semantic relationship between these information and convert it into a pattern that can be learned by the computer, in the technical solution of the present application, semantic embedding encoding is performed on the basic information of the user to map the text to a low-dimensional vector space, mining the potential semantic association in the user's basic information, and obtaining a user basic information semantic embedding encoding feature vector. This helps to more comprehensively understand the user's characteristics and provide richer information for subsequent user group identification.

[0055] Specifically, an embodiment of semantic embedding encoding can be achieved by using a pre-trained language model. In one example, when a user inputs a natural language description of personalized meteorological information service requirements, the system first collects basic information related to the user, such as the user's age, occupation, hobbies, permanent geographical location, and past behavior records of using meteorological services. This data is integrated into a detailed description text about the user. Next, in order to transform this complex text information into a form that can be understood and processed by a computer, the system uses a pre-trained language model (such as BERT, Word2Vec, or GloVe, etc.) to encode the above text. This process is not just a simple keyword matching or rule-based method, but deeply explores the context relationships between words, so that each word is represented as a vector in a multi-dimensional space, and the entire sentence or paragraph is composed of a series of such vectors, forming a feature matrix. This representation method allows the system to capture deeper semantic information, such as the similarity between words, syntactic structures, and contextual implications. Taking BERT as an example, it is a bidirectional encoder representation model that can learn rich language features from a large amount of unlabeled text. In this case, BERT will accept the user's basic information as input and output the corresponding low-dimensional vector representation. Since BERT considers the left and right contexts of words in a sentence, it can better understand the meaning differences of a word like "farmer" in different scenarios - for an agricultural worker, "rainfall" may mean an irrigation water source; while for urban residents, "rainfall" may be a factor affecting travel. The semantic embedding encoding feature vector of the user's basic information generated in this way not only contains static personal information but also integrates dynamic and contextual user preferences and behavior patterns. Finally, these feature vectors after semantic embedding encoding will be used for subsequent user group identification, feature description, and interaction analysis with personalized meteorological information service requirements, so as to provide more accurate and personalized meteorological service recommendation content. This method greatly improves the intelligence level of the meteorological service system, enabling it to provide customized suggestions according to the unique needs of users and enhancing the user experience.

[0056] Next, in order to divide users into different groups and provide customized services according to the characteristics and needs of different groups, for example, providing customized meteorological service information for different groups such as outdoor sports enthusiasts, farmers, and urban residents, the present application inputs the semantic embedded coding feature vector of the user's basic information into a classifier-based user group identifier to obtain the user group identification result. In particular, the classifier is a mature technology in the field of machine learning that has been verified through long-term research and practice. It has demonstrated powerful classification capabilities in many fields, such as image recognition, speech recognition, and text classification. For the user group identification task, the classifier can use its mature algorithms and model structures to efficiently process the encoded user feature vectors, extract key information from complex vector data, and thus achieve accurate classification of user groups.

[0057] Correspondingly, in step S120, performing user group identification on the user based on the user's basic information to obtain a user group identification result includes: performing semantic embedded coding on the user's basic information to obtain a semantic embedded coding feature vector of the user's basic information; and inputting the semantic embedded coding feature vector of the user's basic information into a classifier-based user group identifier to obtain the user group identification result.

[0058] Then, considering that although the user group identification result can determine the group to which the user belongs, these labels themselves may lack detailed descriptions of group characteristics. Based on this, in the technical solution of the present application, the user group identification result is input into a user group feature descriptor based on a large language model to utilize the powerful language generation ability of the large language model to generate coherent, accurate, and rich text descriptions according to the understood content, and obtain a user group feature description text. That is, the large language model has been trained on a large-scale corpus and has extensive language understanding and generation capabilities. They can extract relevant information from existing knowledge and provide more abundant and detailed descriptions of user group characteristics, including but not limited to hobbies, behavior patterns, demand preferences, etc. By generating description texts through the large language model, the commonalities and individual differences of each group can be captured more comprehensively. For example, for the group of seafood farmers in the southern coastal areas, the large language model can combine various information such as local climate characteristics, fishery culture, and market demand to generate a comprehensive and accurate feature description text, thereby providing rich semantic support for subsequent personalized services.

[0059] Correspondingly, in step S130, performing user feature description on the user group identification result to obtain a user group feature description text includes: inputting the user group identification result into a user group feature descriptor based on a large language model to obtain the user group feature description text.

[0060] Next, in order to further extract the semantic information in the described user group feature description text and the natural language description of the personalized meteorological information service requirements, so as to better understand and process the semantic content in the text, this application converts the text into numerical vectors in a high-dimensional space by performing semantic encoding on the described user group feature description text and the natural language description of the personalized meteorological information service requirements respectively, enabling the machine to understand and process these complex semantic relationships, and obtaining the user group feature description semantic encoding vector and the personalized meteorological information service requirements semantic encoding vector. In particular, in a specific embodiment of this application, a semantic encoder based on Bert-Bidirectional LSTM is used to perform semantic encoding on the described user group feature description text and the natural language description of the personalized meteorological information service requirements respectively. It should be understood that Bert is based on the self-attention structure of Transformer and has the ability of parallelized training, which can greatly improve the training efficiency. When processing the user group feature description text and the natural language description of the personalized meteorological information service requirements, it can quickly grasp the overall text semantics and capture the global information of the text sequence. Although Bert has many advantages, the encoding layer based on the self-attention structure is difficult to effectively obtain the order and grammatical relationship between words in the text sequence, and the effect of position embedding is also not satisfactory. As a special recurrent neural network, Bidirectional LSTM can solve the gradient problem in the processing of long sequences, and by processing the sequence bidirectionally, it is especially good at capturing the order and grammatical relationship between words, and can further mine the context details and order information at the word level for the text processed by Bert. Therefore, by processing with a semantic encoder based on Bert-Bidirectional LSTM, the advantages of both can be combined to generate more representative and distinguishable semantic encoding vectors.

[0061] Correspondingly, in step S140, semantic encoding is performed on the described user group feature description text and the natural language description of the personalized meteorological information service requirements respectively to obtain the user group feature description semantic encoding feature and the personalized meteorological information service requirements semantic encoding feature, including: using a semantic encoder based on Bert-Bidirectional LSTM to perform semantic encoding on the described user group feature description text and the natural language description of the personalized meteorological information service requirements respectively to obtain the user group feature description semantic encoding vector as the user group feature description semantic encoding feature and the personalized meteorological information service requirements semantic encoding vector as the personalized meteorological information service requirements semantic encoding feature.

[0062] Furthermore, considering that the semantic encoding features of the user group characteristics description contain the semantic information of the characteristics description of the user's group in terms of meteorological service needs, and the semantic encoding features of the personalized meteorological information service needs express the specific needs of the user for specific meteorological activities. There are complex mutual interaction relationships and associations between these two. Therefore, in order to deeply explore the hidden semantic associations between these two vectors, the present application introduces a bidirectional semantic association attention interaction mechanism to interact with the semantic encoding features of the user group characteristics description and the semantic encoding features of the personalized meteorological information service needs. In particular, this mechanism allows considering the bidirectional influence between the two vectors during the interaction. The characteristics of the user group may affect the focus of personalized needs, and conversely, personalized needs can also refine and specify the general needs contained in the user group characteristics, so as to generate user group characteristics - personalized needs semantic interaction encoding features that are comprehensive and can accurately reflect the actual needs of the user in the meteorological service scenario.

[0063] Correspondingly, in step S150, as Figure 2 shown, performing bidirectional semantic association attention interaction on the semantic encoding features of the user group characteristics description and the semantic encoding features of the personalized meteorological information service needs to obtain user group characteristics - personalized needs semantic interaction encoding features, including: S151, performing a homography projection transformation on the semantic encoding features of the user group characteristics description and the semantic encoding features of the personalized meteorological information service needs to obtain a homography projection encoding feature of the user group characteristics description and a homography projection encoding feature of the personalized meteorological information service needs; S152, performing a bidirectional attention field modulation response on the homography projection encoding feature of the user group characteristics description and the homography projection encoding feature of the personalized meteorological information service needs to obtain the user group characteristics - personalized needs semantic interaction encoding features.

[0064] Among them, in step S151, performing a homography projection transformation on the semantic encoding features of the user group characteristics description and the semantic encoding features of the personalized meteorological information service needs to obtain a homography projection encoding feature of the user group characteristics description and a homography projection encoding feature of the personalized meteorological information service needs, including: using a homography matrix of the user group characteristics description mapping to perform a homography projection transformation on the semantic encoding vector of the user group characteristics description to obtain a homography projection encoding feature vector of the user group characteristics description as the homography projection encoding feature of the user group characteristics description; using a homography matrix of the personalized meteorological information service needs mapping to perform a homography projection transformation on the semantic encoding vector of the personalized meteorological information service needs to obtain a homography projection encoding feature vector of the personalized meteorological information service needs as the homography projection encoding feature of the personalized meteorological information service needs.

[0065] Among them, in step S152, asFigure 3 As shown, a bidirectional attention field modulation response is performed on the homographic projection coding feature of the user group feature description and the homographic projection coding feature of the personalized meteorological information service demand to obtain the user group feature-personalized demand semantic interaction coding feature, including: S1521, constructing a positive and negative user group feature-personalized demand bidirectional attention balance field between the homographic projection coding feature vector of the user group feature description and the homographic projection coding feature vector of the personalized meteorological information service demand; S1522, mapping the homographic projection coding feature vector of the user group feature description and the homographic projection coding feature vector of the personalized meteorological information service demand to the positive and negative user group feature-personalized demand bidirectional attention balance field respectively to obtain a homographic projection attention semantic modulation coding vector of the user group feature description and a homographic projection attention semantic modulation coding vector of the personalized meteorological information service demand; S1523, calculating the division of the homographic projection attention semantic modulation coding vector of the user group feature description and the homographic projection attention semantic modulation coding vector of the personalized meteorological information service demand at each position point to obtain a user group feature-personalized demand semantic interaction coding vector as the user group feature-personalized demand semantic interaction coding feature. Specifically, constructing a positive and negative user group feature-personalized demand bidirectional attention balance field between the homographic projection coding feature vector of the user group feature description and the homographic projection coding feature vector of the personalized meteorological information service demand includes: calculating a positive user group feature-personalized demand attention score field of the homographic projection coding feature vector of the user group feature description relative to the homographic projection coding feature vector of the personalized meteorological information service demand; calculating a negative user group feature-personalized demand attention score field of the homographic projection coding feature vector of the personalized meteorological information service demand relative to the homographic projection coding feature vector of the user group feature description; performing feature splicing and convolutional kernel coding on the positive user group feature-personalized demand attention score field and the negative user group feature-personalized demand attention score field to obtain the positive and negative user group feature-personalized demand bidirectional attention balance field.

[0066] First, a homographic projection transformation is performed on the original user group feature description semantic coding vector and the personalized meteorological information service demand semantic coding vector to map the user group feature description semantic coding vector and the personalized meteorological information service demand semantic coding vector into a new feature space. In this feature space, the relative position relationship between features is maintained, but these features may be observed from different perspectives or scales.

[0067] This process can be represented by the formula:

[0068] v′ a = H1(va )

[0069] v' b = H2(v b )

[0070] where v a is the semantic encoding vector of the user group characteristic description, v b is the semantic encoding vector of the personalized meteorological information service demand, H1 is the homography matrix of the user group characteristic description mapping, H2 is the homography matrix of the personalized meteorological information service demand mapping, and v' a is the homography projection encoding feature vector of the user group characteristic description, and v' b is the homography projection encoding feature vector of the personalized meteorological information service demand.

[0071] Next, calculate the forward user group characteristic - personalized demand attention score field for the encoding vector after the homography projection transformation. This process evaluates the importance distribution of the homography projection encoding feature vector of the user group characteristic description relative to the homography projection encoding feature vector of the personalized meteorological information service demand. It usually involves calculating the position - sensitive similarity score between the two encoding vectors to construct an importance map representing the semantic encoding vector of the user group characteristic description in the context of the semantic encoding vector of the personalized meteorological information service demand. The core of the forward attention mechanism is to quantify the correlation between these two encoding vectors. Such a mechanism mimics the characteristics of the human visual system, enabling the model to "focus" on the part of the semantic encoding vector of the user group characteristic description that best reflects the correlation with the semantic encoding vector of the personalized meteorological information service demand, and at the same time acting as a filter to filter out irrelevant noise information.

[0072] This process can be expressed by the formula:

[0073]

[0074] where v' a is the homography projection encoding feature vector of the user group characteristic description, v' b is the homography projection encoding feature vector of the personalized meteorological information service demand, v' b T is the transposed vector of v' b , is matrix multiplication, S1 is the length of v' b , and M z is the forward user group characteristic - personalized demand attention score field.

[0075] To supplement the information of the positive user group feature - personalized demand attention score field, the reverse user group feature - personalized demand attention score field of the personalized meteorological information service demand homography projection coding feature vector relative to the user group feature description homography projection coding feature vector is then calculated. This process provides an opportunity to understand the relationship between features from the opposite direction, ensuring that important interaction details are not overlooked. In fact, the positive user group feature - personalized demand attention score field and the reverse user group feature - personalized demand attention score field work together to form a more complete and detailed relationship map, which helps to capture asymmetric relationships.

[0076] This process can be expressed by the formula:

[0077]

[0078] where v′ a is the user group feature description homography projection coding feature vector, v′ b is the personalized meteorological information service demand homography projection coding feature vector, v′ a T is the transposed vector of v′ a , is matrix multiplication, S2 is the length of v′ a and M f is the reverse user group feature - personalized demand attention score field.

[0079] To integrate the information of the positive user group feature - personalized demand attention score field and the reverse user group feature - personalized demand attention score field, a positive and negative user group feature - personalized demand bidirectional attention balance field is constructed. This operation comprehensively considers the mutual influence between the two feature vectors (the user group feature description homography projection coding feature vector and the personalized meteorological information service demand homography projection coding feature vector), ensuring that neither side is overly biased nor any side's contribution is ignored in the final interaction response coding. The design of the positive and negative user group feature - personalized demand bidirectional attention balance field needs to consider the specific requirements of the task. Dynamically adjusting the weight coefficients can automatically optimize the balance effect according to the characteristics of the input data, thereby improving the adaptability and generalization ability of the model. In an ideal situation, the positive and negative bidirectional attention should complement each other rather than cancel each other out.

[0080] This process can be expressed by the formula:

[0081] Ω = Conv 3×3 [cat(M z , M f )]

[0082] where M zis the positive user group feature - personalized demand attention score field, M f is the negative user group feature - personalized demand attention score field, cat is the feature concatenation operation, Conv 3×3 is the convolution encoding with a convolution kernel of 3×3, and Ω is the positive and negative user group feature - personalized demand bidirectional attention balance field.

[0083] Next, map the user group feature description homographic projection encoding feature vector and the personalized meteorological information service demand homographic projection encoding feature vector to the positive and negative user group feature - personalized demand bidirectional attention balance field respectively to obtain the user group feature description homographic projection attention semantic modulation encoding vector and the personalized meteorological information service demand homographic projection attention semantic modulation encoding vector. It should be understood that when the user group feature description homographic projection encoding feature vector and the personalized meteorological information service demand homographic projection encoding feature vector are respectively mapped to the previously constructed positive and negative user group feature - personalized demand bidirectional attention balance field, it is actually performing a re - weighting operation. The modulated encoding vectors (the user group feature description homographic projection encoding feature vector and the personalized meteorological information service demand homographic projection encoding feature vector) not only contain the information of the original features but also incorporate the influence from the features of the other party. This process is similar to the feedback mechanism in human communication - one party will adjust their expression according to the reaction of the other party. In this way, the model can understand and represent the complex interactions between features at a higher abstract level, ensuring that each of the user group feature description homographic projection encoding feature vector and the personalized meteorological information service demand homographic projection encoding feature vector adjusts itself according to the guidance provided by the positive and negative user group feature - personalized demand bidirectional attention balance field to generate new encoding vectors - namely, the user group feature description homographic projection attention semantic modulation encoding vector and the personalized meteorological information service demand homographic projection attention semantic modulation encoding vector.

[0084] This process can be expressed by the formula:

[0085]

[0086] where, v′ a is the user group feature description homographic projection encoding feature vector, v′ b is the personalized meteorological information service demand homographic projection encoding feature vector, is matrix multiplication, Ω is the positive and negative user group feature - personalized demand bidirectional attention balance field, v at is the user group feature description homographic projection attention semantic modulation encoding vector, v bt is the personalized meteorological information service demand homographic projection attention semantic modulation encoding vector.

[0087] Finally, by performing element-wise division on the homographic projection attention semantic modulation encoding vector of the user group characteristic description and the homographic projection attention semantic modulation encoding vector of the personalized meteorological information service demand, the final user group characteristic-personalized demand semantic interaction encoding vector is obtained. The division operation is used here as a comparison mechanism, which can reveal the new relationship between the two feature vectors (the homographic projection attention semantic modulation encoding vector of the user group characteristic description and the homographic projection attention semantic modulation encoding vector of the personalized meteorological information service demand) after their respective modulations, providing an informative representation for subsequent tasks. This deep feature contrast and integration reveals the subtle proportional relationships between the features, which are crucial for interpreting the model's decisions.

[0088] This process can be represented by the formula:

[0089]

[0090] where, v at is the homographic projection attention semantic modulation encoding vector of the user group characteristic description, v bt is the homographic projection attention semantic modulation encoding vector of the personalized meteorological information service demand, and V i is the user group characteristic-personalized demand semantic interaction encoding vector.

[0091] Subsequently, after adding a prompt word to the tail of the user group characteristic-personalized demand semantic interaction encoding vector, it is input into the meteorological service key point information generator based on the large language model to obtain the recommended content of the meteorological service key point information prompt. It should be understood that although the large language model has powerful language generation capabilities, its output has a certain degree of openness and uncertainty when facing specific tasks. Adding a prompt word can clarify the task orientation for the large language model and focus its generated content on the category of meteorological service key point information. For example, if the prompt word is "generate the key points of meteorological services suitable for mountaineering activities", the large language model can, based on this guidance, start from the information contained in the user group characteristic-personalized demand semantic interaction encoding vector and generate meteorological key points closely related to mountaineering activities, such as the temperature change in the mountain area and the impact of wind on the mountaineering path, etc., to provide comprehensive meteorological information support for the user's activities.

[0092] Accordingly, in step S160, based on the user group feature-personalized demand semantic interaction coding feature, the recommended content for meteorological service key point information prompt is obtained, including: after adding a prompt word to the tail of the user group feature-personalized demand semantic interaction coding vector, inputting it into the meteorological service key point information generator based on a large language model to obtain the recommended content for the meteorological service key point information prompt.

[0093] Preferably, considering that the user group feature description semantic coding vector and the personalized meteorological information service demand semantic coding vector respectively represent the text semantic coding features of user group features and the text semantic coding features of personalized meteorological information service demands, when performing semantic association interaction based on the positive and negative attention fields of features, the complexity of the positive and negative attention field mechanisms caused by different coding modes of heterogeneous text data will lead to insufficient long-distance interaction response representation of the user group feature-personalized demand semantic interaction coding vector, thereby reducing the expression effect of the user group feature-personalized demand semantic interaction coding vector and affecting the accuracy of the recommended content of the meteorological service key point information prompt generated by inputting it into the meteorological service key point information generator based on a large language model.

[0094] Therefore, in one example, when inputting the user group feature-personalized demand semantic interaction coding vector into the meteorological service key point information generator based on a large language model after adding a prompt word to its tail, the user group feature-personalized demand semantic interaction coding vector is optimized, and the optimization includes the following steps:

[0095] Arrange the respective feature values of the user group feature-personalized demand semantic interaction coding vector in ascending order to obtain the user group feature-personalized demand semantic interaction sequential coding vector;

[0096] In response to the absolute value of the difference between the i-th feature value and the (i + 1)-th feature value of the user group feature-personalized demand semantic interaction sequential coding vector being less than or equal to the distance difference hyperparameter ε, calculate the weighted sum of the i-th feature value and the (i + 1)-th feature value as the optimized (i + 1)-th feature value v′ i+1 = ω1 × v i + ω2 × v i+1 ; where, v i and v i+1 respectively represent the i-th feature value and the (i + 1)-th feature value of the user group feature-personalized demand semantic interaction sequential coding vector, ω1 and ω2 represent weighted hyperparameters, and v′ i+1 represents the optimized (i + 1)-th feature value;

[0097] In response to the absolute value of the difference between the i-th eigenvalue and the (i + 1)-th eigenvalue of the user group feature-personalized demand semantic interaction order coding vector being greater than the distance difference hyperparameter ε, calculate the square root of the sum of the squares of all eigenvalues of the user group feature-personalized demand semantic interaction coding vector where v1 2 + v2 2 + … represents the sum of the squares of all eigenvalues of the user group feature-personalized demand semantic interaction coding vector, and root represents the square root;

[0098] Multiply the square root root by 2 and then divide by the square of the length of the user group feature-personalized demand semantic interaction coding vector to obtain the user group feature-personalized demand semantic interaction space primitive value base = 2×root / L 2 ; where root represents the square root, L represents the length of the user group feature-personalized demand semantic interaction coding vector, and base represents the user group feature-personalized demand semantic interaction space primitive value;

[0099] After multiplying the user group feature-personalized demand semantic interaction space primitive value by the i-th eigenvalue, calculate the weighted subtraction between the product and the (i + 1)-th eigenvalue to obtain the optimized (i + 1)-th eigenvalue v i+1 ′ = ω3×v i ×base - ω4×v i+1 ; where base represents the user group feature-personalized demand semantic interaction space primitive value, v i and v i+1 respectively represent the i-th eigenvalue and the (i + 1)-th eigenvalue of the user group feature-personalized demand semantic interaction order coding vector, ω3 and ω4 represent weighted hyperparameters, and v i+1 ′ represents the optimized (i + 1)-th eigenvalue;

[0100] Based on v′1 = v1, combine the optimized (i + 1)-th eigenvalue v i+1 ′ to obtain the optimized user group feature-personalized demand semantic interaction coding vector; where v1 represents the first eigenvalue of the user group feature-personalized demand semantic interaction order coding vector, and v′1 represents the optimized first eigenvalue.

[0101] In this way, for the problem of insufficient global interaction response representation ability caused by long distances exceeding the predetermined local distribution interval threshold in the feature set of the user group feature-personalized demand semantic interaction coding vector under the predetermined eigenvalue order distribution, the high-dimensional feature space primitive representation based on self-inner product fusion of the user group feature-personalized demand semantic interaction coding vector is used to capture the complex structure of the eigenvalue global network interaction, so as to reconstruct the interaction response relationship between eigenvalues of the user group feature-personalized demand semantic interaction coding vector by simulating the potential primitive of the high-dimensional feature space based on scale, so as to realize the coding reconstruction of the real sequence distribution behavior of the user group feature-personalized demand semantic interaction coding vector under long distances, improve the coding expression effect of the user group feature-personalized demand semantic interaction coding vector, and improve the accuracy of the recommended content of the meteorological service key point information prompt generated by inputting it into the meteorological service key point information generator based on the large language model.

[0102] Based on the above embodiments, refer to Figure 4 As shown, it is a schematic structural diagram of a meteorological service key point information prompt system 100 in an embodiment of the present application. The meteorological service key point information prompt system 100 includes: a demand acquisition module 110, configured to acquire a natural language description of the personalized meteorological information service demand input by the user; a user group identification module 120, configured to extract the basic information of the user and perform user group identification on the user based on the basic information of the user to obtain a user group identification result; a user feature description module 130, configured to perform user feature description on the user group identification result to obtain a user group feature description text; a semantic coding module 140, configured to perform semantic coding on the user group feature description text and the natural language description of the personalized meteorological information service demand respectively to obtain a user group feature description semantic coding feature and a personalized meteorological information service demand semantic coding feature; a bidirectional semantic association attention interaction module 150, configured to perform bidirectional semantic association attention interaction on the user group feature description semantic coding feature and the personalized meteorological information service demand semantic coding feature to obtain a user group feature-personalized demand semantic interaction coding feature; a recommendation module 160, configured to obtain recommended content of meteorological service key point information prompts based on the user group feature-personalized demand semantic interaction coding feature.

[0103] Among them, the user group identification module 120 includes: a semantic embedding coding unit, configured to perform semantic embedding coding on the basic information of the user to obtain a user basic information semantic embedding coding feature vector; a user group identification unit, configured to input the user basic information semantic embedding coding feature vector into a user group identification based on a classifier to obtain the user group identification result.

[0104] Among them, the bidirectional semantic association attention interaction module 150 includes: a homography projection transformation unit, configured to perform homography projection transformation on the user group feature description semantic encoding feature and the personalized meteorological information service demand semantic encoding feature to obtain a user group feature description homography projection encoding feature and a personalized meteorological information service demand homography projection encoding feature; a bidirectional attention field modulation response unit, configured to perform bidirectional attention field modulation response on the user group feature description homography projection encoding feature and the personalized meteorological information service demand homography projection encoding feature to obtain the user group feature-personalized demand semantic interaction encoding feature.

[0105] Here, those skilled in the art can understand that the specific functions and operations of each module in the above meteorological service key point information prompt system 100 have been described in detail above with reference to Figures 1 to 3 the description of the meteorological service key point information prompt method, and therefore, the repeated description thereof will be omitted.

[0106] Those of ordinary skill in the art can understand that all or part of the steps in the above method can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a magnetic disk, or an optical disc, etc. Optionally, all or part of the steps of the above embodiments can also be implemented using one or more integrated circuits. Correspondingly, each module / unit in the above embodiments can be implemented in the form of hardware or in the form of a software functional module. The present application is not limited to any specific form of combination of hardware and software.

[0107] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those of ordinary skill in the art to which this application belongs. It should also be understood that terms such as those defined in a general dictionary should be interpreted as having a meaning consistent with their meaning in the context of the relevant art, and should not be interpreted in an idealized or overly formal sense unless explicitly defined as such herein.

[0108] The above is the description of the present application and should not be regarded as a limitation thereof. Although several exemplary embodiments of the present application have been described, those skilled in the art will easily understand that many modifications can be made to the exemplary embodiments without departing from the novel teachings and advantages of the present application.

Claims

1. A method for prompting key information of meteorological services, characterized in that: include: Obtaining a natural language description of personalized meteorological information service requirements input by a user; Extracting basic information of the user, and performing user group identification on the user based on the basic information of the user to obtain a user group identification result; Performing user feature description on the user group identification result to obtain a user group feature description text; Semantically encoding the user group feature description text and the natural language description of the personalized meteorological information service demand respectively to obtain semantic encoding features of the user group feature description and semantic encoding features of the personalized meteorological information service demand; A bidirectional semantic association attention interaction is performed on the semantic coding features of the user group feature description and the semantic coding features of the personalized meteorological information service requirements to obtain user group feature-personalized demand semantic interaction coding features, including: performing homography projection transformation on the semantic coding features of the user group feature description and the semantic coding features of the personalized meteorological information service requirements to obtain user group feature description homography projection coding features and personalized meteorological information service requirements homography projection coding features; performing bidirectional attention field modulation response on the homography projection coding features of the user group feature description and the homography projection coding features of the personalized meteorological information service requirements to obtain the user group feature-personalized demand semantic interaction coding features; Based on the user group characteristics-personalized demand semantic interaction coding characteristics, the recommended content of the meteorological service key information prompts is obtained.

2. The method for prompting key information of meteorological services according to claim 1, characterized in that: Performing user group identification on the user based on the basic information of the user to obtain a user group identification result includes: Performing semantic embedding coding on the basic information of the user to obtain a semantic embedding coding feature vector of the basic information of the user; The user basic information semantic embedding encoding feature vector is input into the classifier-based user group identification to obtain the user group identification result.

3. The method for prompting key information of meteorological services according to claim 2, characterized in that: Performing user feature description on the user group identification result to obtain user group feature description text includes: inputting the user group identification result into a user group feature descriptor based on a large language model to obtain the user group feature description text.

4. The method for prompting key information of meteorological services according to claim 3, characterized in that: Semantically encode the user group feature description text and the natural language description of the personalized meteorological information service demand respectively to obtain user group feature description semantic coding features and personalized meteorological information service demand semantic coding features, including: using a Bert-bidirectional LSTM-based semantic encoder to semantically encode the user group feature description text and the natural language description of the personalized meteorological information service demand respectively to obtain a user group feature description semantic coding vector as the user group feature description semantic coding feature and a personalized meteorological information service demand semantic coding vector as the personalized meteorological information service demand semantic coding feature.

5. The method for prompting key information of meteorological services according to claim 4, characterized in that: Performing homography projection transformation on the semantic coding features of the user group feature description and the semantic coding features of the personalized meteorological information service requirements to obtain homography projection coding features of the user group feature description and homography projection coding features of the personalized meteorological information service requirements, including: Using the user group feature description mapping homography matrix to perform homography projection transformation on the user group feature description semantic coding vector to obtain a user group feature description homography projection coding feature vector as the user group feature description homography projection coding feature; The personalized meteorological information service demand mapping homography matrix is ​​used to perform homography projection transformation on the personalized meteorological information service demand semantic coding vector to obtain a personalized meteorological information service demand homography projection coding feature vector as the personalized meteorological information service demand homography projection coding feature.

6. The method for prompting key information of meteorological services according to claim 5, characterized in that: A bidirectional attention field modulation response is performed on the homography projection coding feature describing the user group characteristics and the homography projection coding feature of the personalized meteorological information service demand to obtain the user group characteristics-personalized demand semantic interaction coding feature, including: Constructing a positive and negative user group feature-personalized demand bidirectional attention balance field between the user group feature description homography projection coding feature vector and the personalized meteorological information service demand homography projection coding feature vector; Mapping the user group feature description homography projection coding feature vector and the personalized meteorological information service demand homography projection coding feature vector to the positive and negative user group feature-personalized demand bidirectional attention balance field respectively to obtain the user group feature description homography projection attention semantic modulation coding vector and the personalized meteorological information service demand homography projection attention semantic modulation coding vector; Calculate the user group feature description homography projection attention semantic modulation coding vector and the personalized meteorological information service demand homography projection attention semantic modulation coding vector by position point and divide them to obtain the user group feature-personalized demand semantic interaction coding vector as the user group feature-personalized demand semantic interaction coding feature.

7. The method for prompting key information of meteorological services according to claim 6, characterized in that: Constructing a positive and negative user group feature-personalized demand bidirectional attention balance field between the user group feature description homography projection coding feature vector and the personalized meteorological information service demand homography projection coding feature vector, including: Calculate the positive user group feature-personalized demand attention score field of the homography projection coding feature vector of the user group feature description relative to the homography projection coding feature vector of the personalized meteorological information service demand; Calculate the inverse user group feature-personalized demand attention score field of the homography projection coding feature vector of the personalized meteorological information service demand relative to the homography projection coding feature vector of the user group feature description; The positive user group feature-personalized demand attention score field and the reverse user group feature-personalized demand attention score field are subjected to feature splicing and convolution kernel encoding to obtain the positive and reverse user group feature-personalized demand bidirectional attention balance field.

8. The method for prompting key information of meteorological services according to claim 7, characterized in that: Based on the user group characteristics-personalized demand semantic interaction coding features, the recommended content of the meteorological service key information prompt is obtained, including: after adding a prompt word at the end of the user group characteristics-personalized demand semantic interaction coding vector, it is input into a meteorological service key information generator based on a large language model to obtain the recommended content of the meteorological service key information prompt.

9. A meteorological service key information prompt system, characterized in that: include: A demand acquisition module is used to obtain a natural language description of the personalized meteorological information service demand input by the user; A user group identification module, used to extract the basic information of the user, and perform user group identification on the user based on the basic information of the user to obtain a user group identification result; A user feature description module, used to perform user feature description on the user group identification result to obtain a user group feature description text; A semantic coding module, used to semantically code the user group feature description text and the natural language description of the personalized meteorological information service demand to obtain semantic coding features of the user group feature description and semantic coding features of the personalized meteorological information service demand; A bidirectional semantic association attention interaction module, used for performing bidirectional semantic association attention interaction on the semantic coding features of the user group feature description and the semantic coding features of the personalized meteorological information service requirements to obtain user group feature-personalized demand semantic interaction coding features; wherein the bidirectional semantic association attention interaction module includes: a homography projection transformation unit, used for performing homography projection transformation on the semantic coding features of the user group feature description and the semantic coding features of the personalized meteorological information service requirements to obtain user group feature description homography projection coding features and personalized meteorological information service requirements homography projection coding features; a bidirectional attention field modulation response unit, used for performing bidirectional attention field modulation response on the homography projection coding features of the user group feature description and the homography projection coding features of the personalized meteorological information service requirements to obtain the user group feature-personalized demand semantic interaction coding features; The recommendation module is used to obtain recommended content of meteorological service key information prompts based on the user group characteristics-personalized demand semantic interaction coding characteristics.

10. The weather service key information prompt system according to claim 9, characterized in that: The user group identification module includes: A semantic embedding coding unit, used for performing semantic embedding coding on the basic information of the user to obtain a semantic embedding coding feature vector of the user basic information; The user group identification unit is used to input the user basic information semantic embedding coding feature vector into the classifier-based user group identification to obtain the user group identification result.

Images