Information retrieval intelligent dialogue method and system based on large language model

Abstract

The invention discloses an information retrieval intelligent dialogue method and system based on a language large model. The method comprises the steps of data collection, data preprocessing, language model training, information retrieval and system optimization. The invention belongs to the technical field of data processing, and particularly relates to an information retrieval intelligent dialogue method and system based on a large language model.According to the scheme, language model training is adopted, a vocabulary is established, a language model is constructed, a hidden layer, an output layer and time complexity are calculated, a stochastic gradient descent algorithm is used, a loss function is minimized, and recognition accuracy is improved; generating an answer matched with the user demand; according to the method, the optimal decision state function of the web portal information at the specific time is adopted to retrieve the web portal information, the context information in the text data is utilized, the search results are sorted through the relevance between modeling words, the relevance of the search results is improved, the search process is accelerated, and the user experience is improved.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, specifically an intelligent dialogue method and system for information retrieval based on a large language model. Background Technology

[0002] A language-based intelligent dialogue system for information retrieval combines language models and information retrieval technologies. It performs semantic understanding and information matching on user input to quickly retrieve relevant information and provide intelligent responses. The aim is to enhance the intelligence and naturalness of human-computer dialogue, providing faster, more accurate, and personalized information services to help users solve problems and offer a better user experience. However, existing language-based intelligent dialogue systems suffer from technical problems such as insufficient word dependence, low accuracy in language system recognition, and mismatch between the answers generated by the language model and the user's questions or needs. They also present challenges for portal websites in ranking search results from massive amounts of information, prioritizing the most useful information, resulting in slow search processes and negatively impacting user experience. Summary of the Invention

[0003] To address the aforementioned issues and overcome the shortcomings of existing technologies, this invention provides an intelligent dialogue method and system for information retrieval based on a large language model. Addressing the technical problems of insufficient word dependence, low accuracy of language system recognition, and mismatch between language model-generated answers and user-posed questions or needs, this invention employs language model training, establishes a vocabulary, constructs a language model, calculates hidden layers, output layers, and time complexity, and uses a stochastic gradient descent algorithm to minimize the loss function, thereby improving the accuracy of language system recognition and generating answers that match user needs. Furthermore, addressing the technical problem of portal websites struggling to rank search results from massive amounts of information and display the most useful information first, resulting in slow search processes and negatively impacting user experience, this invention utilizes the optimal decision state function of portal website information at a specific time for portal website information retrieval. By leveraging contextual information in text data and modeling the relationships between words, search results are ranked, improving the relevance of search results, accelerating the search process, and enhancing user experience.

[0004] The technical solution adopted by this invention is as follows: The intelligent dialogue method for information retrieval based on a large language model provided by this invention includes the following steps:

[0005] Step S1: Data collection, specifically determining the data source and the methods and channels for data collection;

[0006] Step S2: Data preprocessing, specifically data filtering, classification, and labeling;

[0007] Step S3: Language model training, specifically, involves building a vocabulary, constructing a language model, calculating the hidden layer, output layer, and time complexity, and using the stochastic gradient descent algorithm to minimize the loss function;

[0008] Step S4: Information retrieval, specifically, obtaining the information data layer of the portal website, calculating the state function of different retrieval decisions, and obtaining the optimal decision state function of the portal website information at a specific time by summing different time parameters and different decision parameters, combined with relevant weights and decision probabilities, and then performing portal website information retrieval.

[0009] Step S5: System optimization, specifically, evaluating the system's performance, accuracy, and user experience, and implementing corresponding improvement measures.

[0010] Furthermore, in step S1, the data collection specifically involves determining the data source and the methods and channels for data acquisition, including the Internet, partners, and sensors, to ensure the reliability and diversity of the data sources.

[0011] Further, in step S2, the data preprocessing includes the following steps:

[0012] Step S21: Data screening and cleaning. The collected data is screened and filtered. Invalid, duplicate and erroneous data are cleaned and removed to ensure the quality and accuracy of the data.

[0013] Step S22: Data classification. Classify and organize the data according to its type, content, and attributes, and establish appropriate data structures and patterns to facilitate subsequent processing and analysis.

[0014] Step S23: Data labeling, indexing and labeling the data, including adding keywords, tags and timestamps to facilitate retrieval and processing.

[0015] Further, in step S3, the language model training includes the following steps:

[0016] Step S31: Establish a vocabulary to convert text data into a machine-understandable form. To establish the vocabulary, the preprocessed data is encoded to obtain a unique index for each word.

[0017] Step S32: Construct a language model for understanding and generating text. The network structure of the language model includes an input layer, a hidden layer, and an output layer. The training corpus is preprocessed. The input layer receives the encoded text data and passes the information from the input layer to the hidden layer according to the weights. The output of the input layer is calculated using the following formula:

[0018] ;

[0019] In the formula, x(t) represents the output of the input layer, w(t) represents the weights from the input layer to the hidden layer, and s(t-1) represents the output of the hidden layer at the previous time step;

[0020] Step S33: Calculate the hidden layer and output layer. The hidden layer of an RNN processes the information sent by the input neurons, and the output layer represents the network's output. Neurons in each layer communicate through synapses, and the layers are connected by a weight matrix. The formulas used to calculate the hidden layer and output layer are as follows:

[0021] ;

[0022] ;

[0023] In the formula, i represents the index of the input layer neuron, j represents the index of the hidden layer neuron, k represents the index of the output layer neuron, I1 is the number of input layer neurons, I2 is the number of hidden layer neurons, t represents the time step, used to represent the time dependency of the recurrent neural network, and s j (t) represents the output of neuron j in the hidden layer at time t, f() is the activation function of the hidden layer, g() is the activation function of the output layer, and w ij (t) represents the connection weights from input layer neuron i to hidden layer neuron j, u ji The connection weights s represent the connections from hidden layer neuron j to input layer neuron i. j (t-1) is the output of hidden layer neuron j at the previous time step, w jk y represents the connection weight from hidden layer neuron j to output layer neuron k. k (t) represents the output of neuron k in the output layer;

[0024] Step S34: Calculate the time complexity to perform a preliminary evaluation of the language model's data processing performance. Initialize the weight matrices and biases of the hidden and output layers, and calculate the time complexity of each step during model training and testing. The formula used is as follows:

[0025] O = H × (H + V);

[0026] In the formula, O is the time complexity, H is the size of the hidden layer, and V is the size of the vocabulary.

[0027] Step S35: Calculate the loss function using stochastic gradient descent. Update the parameters based on the gradient of each sentence in the training set D, minimizing the negative log-likelihood. Sentences are treated as word tuples and slot tuples. Optimize the model's predictive ability by minimizing the loss function, gradually improving the prediction accuracy for sentences in the training set. The formula used is as follows:

[0028] ;

[0029] In the formula, the length of each sentence varies depending on the training samples, the window size of the context words is a hyperparameter, L(Θ) is the loss function, defined as the sum of negative log-likelihoods, Θ represents the model parameters, (S, W) are each sentence in the training set D, S represents the slot tuple in the sentence, W represents the word tuple in the sentence, th represents the position of each word in the sentence, TH represents the sentence length, d represents the window size of the context words, and P... Θ This represents the output of the model. This represents the probability that the model will predict the th slot tuple given the size of the context word window.

[0030] Further, in step S4, the information retrieval includes the following steps:

[0031] Step S41: Obtain the information data layer of the portal website and calculate the information data layer of the portal website using the following formula:

[0032] ;

[0033] In the formula, N represents the information data layer of the portal website, that is, the information content carried in the portal website; Tq represents the semantic state parameter dataset of the fast portal information retrieval system, that is, the set of semantic state parameters in the fast portal information retrieval system; B Tq This represents the mapping from the semantic state parameter dataset of the fast portal information retrieval system to the portal information dataset. This indicates the mapping from portal website information status parameters to Tq. denoted by , b represents the probability of the retrieval decision, b represents different decision parameters, and W represents the retrieval time parameter;

[0034] Step S42: Calculate the state functions for different retrieval decisions. By performing unstructured processing on the data related to the differential network entry information, the state functions for different retrieval decisions are obtained. The formula used is as follows:

[0035] ;

[0036] In the formula, W a (t1) represents the state function of the a-th retrieval decision, i.e., the execution power of the retrieval decision, t1 represents time, t0 represents a specific time point, and F a The computational function representing the state function maps input parameters to the execution power of decisions. The sum of parameters is obtained after unstructured processing of data related to the entry information of the differential network. This sum is a linear combination of different inputs. η(a) represents the parameter weights, used to adjust the importance of each input. S aThis represents the a-th data point related to the entry information of the differential network;

[0037] Step S43: Calculate the optimal decision state function for the portal website information. By summing different time parameters and different decision parameters, and combining relevant weights and decision probabilities, the optimal decision state function for the portal website information at a specific time is obtained. The formula used is as follows:

[0038] ;

[0039] In the formula, W + (t1) represents the optimal decision state function for portal website information, indicating the optimal decision state at time t1, B is the total number of decision parameters, ω(t1, b) represents the weight of decision parameter b at time t1, t1 represents different time parameters, and T is the total time. This represents the decision probability calculated based on the state parameter t1 and the decision parameter b. This represents the data associated with the time parameter t1, where η is a weighting parameter used to adjust the influence between the state function and the decision function. Indicates time t1 b The decision state function under the given conditions;

[0040] Step S44: Portal website information retrieval. A Markov model for portal website information retrieval is built using the following formula:

[0041] ;

[0042] In the formula, Let represent the optimal decision function for portal website information retrieval in a Markov model, and let represent the optimal decision function with decision parameter b at time t1. This represents the decision quality calculated based on the time parameter t1 and the decision parameter b. This represents the maximum value of the data associated with the time parameter t1.

[0043] Furthermore, in step S5, the system optimization specifically involves continuous optimization based on user feedback and usage. Through user evaluation and technical indicator monitoring, the system's performance, accuracy, and user experience are assessed, and corresponding improvement measures are implemented to enhance the system's quality and effectiveness.

[0044] The intelligent dialogue system for information retrieval based on a large language model provided by this invention includes a data collection module, a data preprocessing module, a language model training module, an information retrieval module, and a system optimization module;

[0045] The data collection module specifically determines the data source and the methods and channels for data collection;

[0046] The data preprocessing module specifically performs data filtering, classification, and labeling.

[0047] The language model training module specifically involves establishing a vocabulary, constructing a language model, calculating the hidden layer, output layer, and time complexity, and using the stochastic gradient descent algorithm to minimize the loss function.

[0048] The information retrieval module specifically acquires the information data layer of the portal website, calculates the state function of different retrieval decisions, and obtains the optimal decision state function of the portal website information at a specific time by summing different time parameters and different decision parameters, combined with relevant weights and decision probabilities, and then performs portal website information retrieval.

[0049] The system optimization module specifically evaluates the system's performance, accuracy, and user experience, and implements corresponding improvement measures.

[0050] The beneficial results achieved by the present invention using the above solution are as follows:

[0051] (1) To address the technical issues of insufficient word dependence, low accuracy of language system recognition, and mismatch between the answers generated by the language model and the questions or needs raised by users, we adopt language model training, establish a vocabulary, construct a language model, calculate the hidden layer, output layer and time complexity, use the stochastic gradient descent algorithm to minimize the loss function, improve the accuracy of language system recognition, and generate answers that match the user's needs.

[0052] (2) To address the technical problem that portal websites have difficulty sorting search results from massive amounts of information and displaying the most useful information first, resulting in a slow search process and affecting user experience, the optimal decision state function of portal website information at a specific time is adopted to retrieve portal website information. By utilizing the contextual information in the text data and modeling the correlation between words, the search results are sorted to improve the relevance of the search results, accelerate the search process, and improve user experience. Attached Figure Description

[0053] Figure 1 This is a flowchart illustrating the intelligent dialogue method for information retrieval based on a large language model provided by the present invention.

[0054] Figure 2 A schematic diagram of the intelligent dialogue system for information retrieval based on a large language model provided by the present invention;

[0055] Figure 3 This is a flowchart illustrating step S3;

[0056] Figure 4 This is a flowchart illustrating step S4.

[0057] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0058] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0059] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0060] Example 1, see Figure 1 The present invention provides an intelligent dialogue method for information retrieval based on a large language model, which includes the following steps:

[0061] Step S1: Data collection, specifically determining the data source and the methods and channels for data collection;

[0062] Step S2: Data preprocessing, specifically data filtering, classification, and labeling;

[0063] Step S3: Language model training, specifically, involves building a vocabulary, constructing a language model, calculating the hidden layer, output layer, and time complexity, and using the stochastic gradient descent algorithm to minimize the loss function;

[0064] Step S4: Information retrieval, specifically, obtaining the information data layer of the portal website, calculating the state function of different retrieval decisions, and obtaining the optimal decision state function of the portal website information at a specific time by summing different time parameters and different decision parameters, combined with relevant weights and decision probabilities, and then performing portal website information retrieval.

[0065] Step S5: System optimization, specifically, evaluating the system's performance, accuracy, and user experience, and implementing corresponding improvement measures.

[0066] Example 2, see Figure 1This embodiment is based on the above embodiment. In step S1, the data collection specifically involves determining the data source and the methods and channels for data collection, including the Internet, partners, and sensors, to ensure the reliability and diversity of data sources.

[0067] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S2, the data preprocessing includes the following steps:

[0068] Step S21: Data screening and cleaning. The collected data is screened and filtered. Invalid, duplicate and erroneous data are cleaned and removed to ensure the quality and accuracy of the data.

[0069] Step S22: Data classification. Classify and organize the data according to its type, content, and attributes, and establish appropriate data structures and patterns to facilitate subsequent processing and analysis.

[0070] Step S23: Data labeling, indexing and labeling the data, including adding keywords, tags and timestamps to facilitate retrieval and processing.

[0071] Example 4, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S3, the language model training includes the following steps:

[0072] Step S31: Establish a vocabulary to convert text data into a machine-understandable form. To establish the vocabulary, the preprocessed data is encoded to obtain a unique index for each word.

[0073] Step S32: Construct a language model for understanding and generating text. The network structure of the language model includes an input layer, a hidden layer, and an output layer. The training corpus is preprocessed. The input layer receives the encoded text data and passes the information from the input layer to the hidden layer according to the weights. The output of the input layer is calculated using the following formula:

[0074] ;

[0075] In the formula, x(t) represents the output of the input layer, w(t) represents the weights from the input layer to the hidden layer, and s(t-1) represents the output of the hidden layer at the previous time step;

[0076] Step S33: Calculate the hidden layer and output layer. The hidden layer of an RNN processes the information sent by the input neurons, and the output layer represents the network's output. Neurons in each layer communicate through synapses, and the layers are connected by a weight matrix. The formulas used to calculate the hidden layer and output layer are as follows:

[0077] ;

[0078] ;

[0079] In the formula, i represents the index of the input layer neuron, j represents the index of the hidden layer neuron, k represents the index of the output layer neuron, I1 is the number of input layer neurons, I2 is the number of hidden layer neurons, t represents the time step, used to represent the time dependency of the recurrent neural network, and s j (t) represents the output of neuron j in the hidden layer at time t, f() is the activation function of the hidden layer, g() is the activation function of the output layer, and w ij (t) represents the connection weights from input layer neuron i to hidden layer neuron j, u ji The connection weights s represent the connections from hidden layer neuron j to input layer neuron i. j (t-1) is the output of hidden layer neuron j at the previous time step, w jk y represents the connection weight from hidden layer neuron j to output layer neuron k. k (t) represents the output of neuron k in the output layer;

[0080] Step S34: Calculate the time complexity to perform a preliminary evaluation of the language model's data processing performance. Initialize the weight matrices and biases of the hidden and output layers, and calculate the time complexity of each step during model training and testing. The formula used is as follows:

[0081] O = H × (H + V);

[0082] In the formula, O is the time complexity, H is the size of the hidden layer, and V is the size of the vocabulary.

[0083] Step S35: Calculate the loss function using stochastic gradient descent. Update the parameters based on the gradient of each sentence in the training set D, minimizing the negative log-likelihood. Sentences are treated as word tuples and slot tuples. Optimize the model's predictive ability by minimizing the loss function, gradually improving the prediction accuracy for sentences in the training set. The formula used is as follows:

[0084] ;

[0085] In the formula, the length of each sentence varies depending on the training samples, the window size of the context words is a hyperparameter, L(Θ) is the loss function, defined as the sum of negative log-likelihoods, Θ represents the model parameters, (S, W) are each sentence in the training set D, S represents the slot tuple in the sentence, W represents the word tuple in the sentence, th represents the position of each word in the sentence, TH represents the sentence length, d represents the window size of the context words, and P... Θ This represents the output of the model. This represents the probability that the model will predict the th slot tuple given the size of the context word window.

[0086] By performing the above operations, language model training is adopted, a vocabulary is established, a language model is constructed, hidden layers, output layers and time complexity are calculated, and the loss function is minimized using the stochastic gradient descent algorithm, thereby improving the accuracy of language system recognition and generating answers that match user needs. This solves the technical problems of insufficient word dependence, low accuracy of language system recognition, and mismatch between the answers generated by the language model and the questions or needs raised by users.

[0087] Example 5, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S4, the information retrieval includes the following steps:

[0088] Step S41: Obtain the information data layer of the portal website and calculate the information data layer of the portal website using the following formula:

[0089] ;

[0090] In the formula, N represents the information data layer of the portal website, that is, the information content carried in the portal website; Tq represents the semantic state parameter dataset of the fast portal information retrieval system, that is, the set of semantic state parameters in the fast portal information retrieval system; B Tq This represents the mapping from the semantic state parameter dataset of the fast portal information retrieval system to the portal information dataset. This indicates the mapping from portal website information status parameters to Tq. denoted by , b represents the probability of the retrieval decision, b represents different decision parameters, and W represents the retrieval time parameter;

[0091] Step S42: Calculate the state functions for different retrieval decisions. By performing unstructured processing on the data related to the differential network entry information, the state functions for different retrieval decisions are obtained. The formula used is as follows:

[0092] ;

[0093] In the formula, W a (t1) represents the state function of the a-th retrieval decision, i.e., the execution power of the retrieval decision, t1 represents time, t0 represents a specific time point, and F a The computational function representing the state function maps input parameters to the execution power of decisions. The sum of parameters is obtained after unstructured processing of data related to the entry information of the differential network. This sum is a linear combination of different inputs. η(a) represents the parameter weights, used to adjust the importance of each input. S aThis represents the a-th data point related to the entry information of the differential network;

[0094] Step S43: Calculate the optimal decision state function for the portal website information. By summing different time parameters and different decision parameters, and combining relevant weights and decision probabilities, the optimal decision state function for the portal website information at a specific time is obtained. The formula used is as follows:

[0095] ;

[0096] In the formula, W + (t1) represents the optimal decision state function for portal website information, indicating the optimal decision state at time t1, B is the total number of decision parameters, ω(t1, b) represents the weight of decision parameter b at time t1, t1 represents different time parameters, and T is the total time. This represents the decision probability calculated based on the state parameter t1 and the decision parameter b. This represents the data associated with the time parameter t1, where η is a weighting parameter used to adjust the influence between the state function and the decision function. Indicates time t1 b The decision state function under the given conditions;

[0097] Step S44: Portal website information retrieval. A Markov model for portal website information retrieval is built using the following formula:

[0098] ;

[0099] In the formula, Let represent the optimal decision function for portal website information retrieval in a Markov model, and let represent the optimal decision function with decision parameter b at time t1. This represents the decision quality calculated based on the time parameter t1 and the decision parameter b. This represents the maximum value of the data associated with the time parameter t1.

[0100] By performing the above operations, the optimal decision state function of portal website information at a specific time is used to retrieve portal website information. By utilizing the contextual information in the text data and modeling the correlation between words, the search results are ranked, improving the relevance of the search results, accelerating the search process, and improving the user experience. This solves the technical problem that portal websites have difficulty ranking search results from massive amounts of information and displaying the most useful information first, resulting in a slow search process and affecting the user experience.

[0101] Example 6, see Figure 1This embodiment is based on the above embodiment. In step S5, the system optimization specifically involves continuous optimization based on user feedback and usage. Through user evaluation and technical indicator monitoring, the system's performance, accuracy, and user experience are assessed, and corresponding improvement measures are implemented to improve the system's quality and effectiveness.

[0102] Example 7, see Figure 2 Based on the above embodiments, the intelligent dialogue system for information retrieval based on a large language model provided by the present invention includes a data collection module, a data preprocessing module, a language model training module, an information retrieval module, and a system optimization module.

[0103] The data collection module specifically determines the data source and the methods and channels for data collection;

[0104] The data preprocessing module specifically performs data filtering, classification, and labeling.

[0105] The language model training module specifically involves establishing a vocabulary, constructing a language model, calculating the hidden layer, output layer, and time complexity, and using the stochastic gradient descent algorithm to minimize the loss function.

[0106] The information retrieval module specifically acquires the information data layer of the portal website, calculates the state function of different retrieval decisions, and obtains the optimal decision state function of the portal website information at a specific time by summing different time parameters and different decision parameters, combined with relevant weights and decision probabilities, and then performs portal website information retrieval.

[0107] The system optimization module specifically evaluates the system's performance, accuracy, and user experience, and implements corresponding improvement measures.

[0108] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0109] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0110] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. An intelligent dialogue method for information retrieval based on a large language model, characterized by: The method includes the following steps: Step S1: Data collection; Step S2: Data preprocessing; Step S3: Language model training, specifically, involves building a vocabulary, constructing a language model, calculating the hidden layer, output layer, and time complexity, and using the stochastic gradient descent algorithm to minimize the loss function; Step S4: Information retrieval, specifically, obtaining the information data layer of the portal website, calculating the state function of different retrieval decisions, and obtaining the optimal decision state function of the portal website information at a specific time by summing different time parameters and different decision parameters, combined with relevant weights and decision probabilities, and then performing portal website information retrieval. Step S5: System optimization.

2. The intelligent dialogue method for information retrieval based on a large language model according to claim 1, characterized in that: In step S3, the language model training includes the following steps: Step S31: Create a vocabulary list; Step S32: Construct a language model. The network structure of the language model includes an input layer, a hidden layer, and an output layer. Preprocess the training corpus and calculate the output of the input layer using the following formula: ； In the formula, x(t) represents the output of the input layer, w(t) represents the weights from the input layer to the hidden layer, and s(t-1) represents the output of the hidden layer at the previous time step; Step S33: Calculate the hidden layer and output layer using the following formulas: ； ； In the formula, i represents the index of the input layer neuron, j represents the index of the hidden layer neuron, k represents the index of the output layer neuron, I1 is the number of input layer neurons, I2 is the number of hidden layer neurons, t represents the time step, used to represent the time dependency of the recurrent neural network, and s j (t) represents the output of neuron j in the hidden layer at time t, f() is the activation function of the hidden layer, g() is the activation function of the output layer, and w ij (t) represents the connection weights from input layer neuron i to hidden layer neuron j, u ji The connection weights s represent the connections from hidden layer neuron j to input layer neuron i. j (t-1) is the output of hidden layer neuron j at the previous time step, w jk y represents the connection weight from hidden layer neuron j to output layer neuron k. k (t) represents the output of neuron k in the output layer; Step S34: Calculate the time complexity, initialize the weight matrices and biases of the hidden and output layers, and calculate the time complexity of each step during model training and testing. The formula used is as follows: O = H × (H + V); In the formula, O is the time complexity, H is the size of the hidden layer, and V is the size of the vocabulary. Step S35: Calculate the loss function, update the parameters based on the gradient of each sentence in the training set D, and minimize the negative log-likelihood. The formula used is as follows: ； In the formula, the length of each sentence varies depending on the training samples, the window size of the context words is a hyperparameter, L(Θ) is the loss function, defined as the sum of negative log-likelihoods, Θ represents the model parameters, (S, W) are each sentence in the training set D, S represents the slot tuple in the sentence, W represents the word tuple in the sentence, th represents the position of each word in the sentence, TH represents the sentence length, d represents the window size of the context words, and P... Θ This represents the output of the model. This represents the probability that the model will predict the th slot tuple given the size of the context word window.

3. The intelligent dialogue method for information retrieval based on a large language model according to claim 1, characterized in that: In step S4, the information retrieval includes the following steps: Step S41: Obtain the information data layer of the portal website and calculate the information data layer of the portal website using the following formula: ； In the formula, N represents the information data layer of the portal website, that is, the information content carried in the portal website; Tq represents the semantic state parameter dataset of the fast portal information retrieval system, that is, the set of semantic state parameters in the fast portal information retrieval system; B Tq This represents the mapping from the semantic state parameter dataset of the fast portal information retrieval system to the portal information dataset. This indicates the mapping from portal website information status parameters to Tq. denoted by , b represents the probability of the retrieval decision, b represents different decision parameters, and W represents the retrieval time parameter; Step S42: Calculate the state functions for different retrieval decisions. By performing unstructured processing on the data related to the differential network entry information, the state functions for different retrieval decisions are obtained. The formula used is as follows: ； In the formula, W a (t1) represents the state function of the a-th retrieval decision, i.e., the execution power of the retrieval decision, t1 represents time, t0 represents a specific time point, and F a The computational function representing the state function maps input parameters to the execution power of decisions. The sum of parameters is obtained after unstructured processing of data related to the entry information of the differential network. This sum is a linear combination of different inputs. η(a) represents the parameter weights, used to adjust the importance of each input. S a This represents the a-th data point related to the entry information of the differential network; Step S43: Calculate the optimal decision state function for the portal website information. By summing different time parameters and different decision parameters, and combining relevant weights and decision probabilities, the optimal decision state function for the portal website information at a specific time is obtained. The formula used is as follows: ； In the formula, W + (t1) represents the optimal decision state function for portal website information, indicating the optimal decision state at time t1, B is the total number of decision parameters, ω(t1, b) represents the weight of decision parameter b at time t1, t1 represents different time parameters, and T is the total time. This represents the decision probability calculated based on the state parameter t1 and the decision parameter b. This represents the data associated with the time parameter t1, where η is a weighting parameter used to adjust the influence between the state function and the decision function. Indicates time t1 b The decision state function under the given conditions; Step S44: Portal website information retrieval. A Markov model for portal website information retrieval is built using the following formula: ； In the formula, Let represent the optimal decision function for portal website information retrieval in a Markov model, and let represent the optimal decision function with decision parameter b at time t1. This represents the decision quality calculated based on the time parameter t1 and the decision parameter b. This represents the maximum value of the data associated with the time parameter t1.

4. The intelligent dialogue method for information retrieval based on a large language model according to claim 1, characterized in that: In step S1, the data collection specifically involves determining the data source and the methods and channels for data collection.

5. The intelligent dialogue method for information retrieval based on a large language model according to claim 1, characterized in that: In step S2, the data preprocessing includes the following steps: Step S21: Data screening and cleaning, screening and filtering the collected data; Step S22: Data classification, classifying and organizing the data according to its type, content, and attributes; Step S23: Data labeling, indexing and labeling the data.

6. The intelligent dialogue method for information retrieval based on a large language model according to claim 1, characterized in that: In step S5, the system optimization specifically involves continuous optimization based on user feedback and usage. The system's performance, accuracy, and user experience are evaluated through user evaluation and technical indicator monitoring, and corresponding improvement measures are implemented.

7. An intelligent dialogue system for information retrieval based on a large language model, used to implement the intelligent dialogue method for information retrieval based on a large language model as described in any one of claims 1-6, characterized in that: It includes a data collection module, a data preprocessing module, a language model training module, an information retrieval module, and a system optimization module.

8. The intelligent dialogue system for information retrieval based on a large language model according to claim 7, characterized in that: The data collection module specifically determines the data source and the methods and channels for data collection; The data preprocessing module specifically performs data filtering, classification, and labeling. The language model training module specifically involves establishing a vocabulary, constructing a language model, calculating the hidden layer, output layer, and time complexity, and using the stochastic gradient descent algorithm to minimize the loss function. The information retrieval module specifically acquires the information data layer of the portal website, calculates the state function of different retrieval decisions, and obtains the optimal decision state function of the portal website information at a specific time by summing different time parameters and different decision parameters, combined with relevant weights and decision probabilities, and then performs portal website information retrieval. The system optimization module specifically evaluates the system's performance, accuracy, and user experience, and implements corresponding improvement measures.

Images