Man-machine conversation method and system for database index management based on large model

Abstract

The invention discloses a man-machine conversation method and system for database index management based on a large model. Based on the multiple matched statement segments, the similarity between every two matched statement segments is judged, and a non-similar statement segment set and a plurality of similar statement segment sets are obtained; constructing a first search engine containing an index structure on the basis of the non-similar statement segment set and the plurality of similar statement segment sets; in the interaction process, answer data more suitable for interaction statements are found through a deep learning network according to the constructed keyword association graph; detecting the similarity of the answer data to obtain a non-similar statement segment set and a plurality of similar statement segment sets; constructing an index structure according to the non-similar statement segment set and the plurality of similar statement segment sets; the index structure can be used for constructing a professional search engine in a vertical field of an industry, and is subdivision and extension of the search engine.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a human-computer dialogue method and system for performing database index management based on a large model. Background Art

[0002] At present, the application direction of the combination of artificial intelligence and data center is very extensive. It can quickly develop intelligent agents for data analysis based on local large models, and the front end can interact with virtual people or other forms. Large models require a lot of data support, but the general search engine composed of large models and interactive front ends has problems such as large amount of information, inaccurate queries, and insufficient depth. Summary of the invention

[0003] The purpose of the present invention is to provide a human-computer dialogue method and system for database index management based on a large model, so as to solve the above-mentioned problems existing in the prior art.

[0004] In a first aspect, an embodiment of the present invention provides a human-computer dialogue method for managing a database index based on a large model, comprising: Obtaining an interactive statement; the interactive statement is a statement for a user to interact during human-computer interaction; Input the interactive sentences into the keyword neural network, detect the keywords, and obtain n first keywords; By matching the search engine, the sentence segments containing the first keyword are detected to obtain m search sentence segments; Calculate the number of first keywords in the search sentence segments respectively as the number of first keywords; obtain m*n numbers of first keywords corresponding to m search sentence segments; Based on the n first keywords, the m search sentence segments and the corresponding number of first keywords, a pre-match is performed to obtain a plurality of matching sentence segments; the matching sentence segments represent sentence segments that match the answering interactive sentences in the search engine; Based on multiple matching sentence segments, the similarity between the matching sentence segments is determined to obtain a set of dissimilar sentence segments and multiple sets of similar sentence segments; a set of similar sentence segments includes multiple similar sentence segments; the set of dissimilar sentence segments includes matching sentence segments that are dissimilar in pairs; the similar sentence segments represent similar matching sentence segments; Based on the dissimilar sentence segment set and a plurality of similar sentence segment sets, a first search engine including an index structure is constructed.

[0005] Optionally, the pre-matching is performed based on the n first keywords, the m search statement segments and the corresponding number of first keywords to obtain multiple matching statement segments, including: Randomly obtain a first keyword as the third keyword; The first keyword except the third keyword is used as the second keyword to obtain a second keyword set; the number of elements in the second keyword set is n-1; Based on the n first keywords, the m second keyword sets and the corresponding number of first keywords, a graph structure is constructed to obtain a keyword association graph; m search sentence segments correspond to m keyword association graphs; Based on the m keyword association graphs and the m search sentence segments, a plurality of matching sentence segments are obtained; the matching sentence segments represent the search sentence segments associated with the interactive sentence.

[0006] Optionally, constructing a graph structure based on the n first keywords, the m second keyword sets and the corresponding number of first keywords to obtain a keyword association graph includes: Establishing an association relationship between the first keyword and a second keyword in the corresponding second keyword set to obtain a keyword pair; the keyword pair includes a first keyword and a second keyword; One first keyword corresponds to n-1 keyword pairs; n first keywords correspond to n*(n-1) keyword pairs; Add the two first keyword quantities corresponding to the keyword pair to obtain the keyword pair value; n*(n-1) keyword pairs correspond to n*(n-1) keyword pair values; Detecting two keyword pairs in which the first keyword and the second keyword are identical, and associating the two keyword pairs as identical keyword pairs; Average the keyword pair values ​​corresponding to the same keyword pair to obtain the average keyword value; For n*(n-1) keyword pairs, n*(n-1) / 2 average keyword values ​​are obtained; Based on the n first keywords and the corresponding average keyword values, a keyword association graph is obtained.

[0007] Optionally, obtaining a keyword association graph based on the n first keywords and the corresponding average keyword values ​​includes: Use the n first keywords as rows and columns to construct a connected graph; The number of rows and columns of the connected graph is the same, and the first keywords corresponding to the rows and columns are the same; According to the first keyword and the second keyword corresponding to the keyword pair, the average keyword value is filled into the corresponding position in the connected graph to obtain a keyword association graph.

[0008] Optionally, the obtaining of multiple matching sentence segments based on the m keyword association graphs and the m search sentence segments includes: Inputting the keyword association graph into a first matching network to obtain a first matching degree feature; Inputting the first matching degree feature into the first classification network to obtain a first matching degree value; and obtaining m first matching degree values ​​corresponding to the m keyword association graphs; The keyword association graph whose first matching degree value is greater than the matching threshold is used as the matching association graph; The search sentence segment corresponding to the matching association graph is taken as the matching sentence segment.

[0009] Optionally, the first matching network includes a first convolution kernel, a second convolution kernel and a third convolution kernel; The first convolution kernel is a 2*2 two-dimensional convolution kernel; the second convolution kernel is an n*2 two-dimensional convolution kernel; the third convolution kernel is a 2*n two-dimensional convolution kernel; In the direction of rows and columns of the keyword association graph, the first convolution kernel is moved with a step size of 1 for convolution to obtain the first feature graph; In the row direction of the keyword association graph, the second convolution kernel is moved with a step size of 1 to perform convolution to obtain a second eigenvector; In the column direction of the keyword association graph, the third convolution kernel is moved with a step size of 1 to perform convolution to obtain a third eigenvector; The first feature map, the second feature vector and the third feature vector are input into a first neural network to obtain a first matching degree feature.

[0010] Optionally, the method of judging the similarity between the matching sentence segments based on the multiple matching sentence segments to obtain a set of dissimilar sentence segments and a set of similar sentence segments includes: In the matching sentence segment, the position of the keyword is detected to obtain a first key point sequence number; the first key point sequence number represents the sequence number of the keyword in the matching sentence segment sorted according to the word order; According to the first key point sequence number, the keywords corresponding to the matching sentence segments are sorted to obtain a first detection sequence; Inputting the first detection sequences in pairs into the discriminant network to judge the similarity, and obtaining a plurality of first similar sentence segment sets; the first similar sentence segment sets include two similar matching sentence segments; Obtaining an intersection of multiple first similar sentence segment sets to obtain a similar sentence segment set; The matching sentence segments other than the matching sentence segments in the similar sentence segment set are used to construct a dissimilar sentence segment set.

[0011] Optionally, constructing a first search engine including an index structure based on the set of dissimilar sentence segments and a plurality of sets of similar sentence segments includes: The index structure includes a first index structure and a second index structure; Taking a matching sentence segment in the similar sentence segment set as the first data to be answered; Putting the first data to be answered into a first index structure; Using the matching sentence segments in the set of dissimilar sentence segments as the second data to be answered; Putting the second data to be answered into a second index structure; Based on the first index structure and the second index structure, a first search engine including an index structure is constructed.

[0012] Optionally, constructing a first search engine including an index structure based on the first index structure and the second index structure includes: The first search engine outputs the first data to be answered in the first index structure; If the first search engine obtains a second interactive statement; the second interactive statement represents a supplement to the interactive statement; The first search engine will be used as a matching search engine in the second index structure to detect the second interactive sentence and obtain a second similar sentence segment set; the second similar sentence segment set represents similar matching sentence segments corresponding to the second interactive sentence.

[0013] In a second aspect, an embodiment of the present invention provides a human-computer dialogue system for managing database indexes based on a large model, including: An acquisition module is used to acquire interaction statements; the interaction statements are statements that users interact with during human-computer interaction; A keyword detection module, used for inputting the interactive sentence into the keyword neural network, detecting the keyword, and obtaining n first keywords; A search module, used to detect a sentence segment containing the first keyword by matching a search engine, and obtain m search sentence segments; A quantity calculation module is used to calculate the number of first keywords in the search statement segments as the number of first keywords; m search statement segments correspond to obtaining m*n numbers of first keywords; A pre-matching module performs pre-matching based on the n first keywords, the m search sentence segments and the corresponding number of first keywords to obtain a plurality of matching sentence segments; the matching sentence segments represent sentence segments that match the answering interactive sentence in the search engine; A similarity judgment module is used to judge the similarity between the matching sentence segments based on multiple matching sentence segments, and obtain a set of dissimilar sentence segments and multiple sets of similar sentence segments; a similar sentence segment set includes multiple similar sentence segments; the dissimilar sentence segment set includes matching sentence segments that are not similar to each other; the similar sentence segments represent similar matching sentence segments; The search engine construction module is used to construct a first search engine including an index structure based on the non-similar sentence segment set and a plurality of similar sentence segment sets.

[0014] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects: In the present invention, during the interaction process, based on the construction of a keyword association graph, answer data that is more suitable for the interactive sentence is found through a deep learning network. The similarity of the answer data is detected, and a set of dissimilar sentence segments and a plurality of similar sentence segment sets are obtained. An index structure is constructed based on the set of dissimilar sentence segments and a plurality of similar sentence segment sets. The index structure can be used to construct a professional search engine in the vertical field of an industry, which is a subdivision and extension of the search engine. A search engine service model is obtained, through which a search service can be provided for a certain type of information. The search engine service model can search web page libraries and local data. The technical effect of obtaining the data required by the user in a field-oriented manner, processing it, and then sending it to the user can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flow chart of a human-computer dialogue method for performing database index management based on a large model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0016] The present invention will be described in detail below in conjunction with the accompanying drawings.

[0017] Example 1 like Figure 1 As shown, an embodiment of the present invention provides a human-computer dialogue method for database index management based on a large model, the method comprising: S101: Obtaining interactive statements; the question statements are statements used by the user to interact during human-computer interaction.

[0018] Among them, in this embodiment, the interactive statement is "What are the common loss functions in deep learning?"

[0019] S102: Input the interactive sentence into the keyword neural network, detect the keywords, and obtain n first keywords.

[0020] Among them, the two random first keywords are different keywords.

[0021] Among them, the keyword neural network is a fully connected neural network (Fully Connected Neural Network, FCN).

[0022] Among them, the detected keywords are keywords stored in the database. For example, if the interactive statement is "What are the common loss functions in deep learning", the database stores keywords such as "deep learning", "common", "loss function", and "include". Then n corresponding to the first keyword is 4, and the 4 first keywords are "deep learning", "common", "loss function", and "include". The keywords stored in the database are used as annotation data to train the keyword neural network.

[0023] Wherein, n is a positive integer.

[0024] S103: Detect the sentence segments containing the first keyword by matching the search engine, and obtain m search sentence segments.

[0025] Wherein, m is a positive integer.

[0026] Among them, one search statement segment contains n first keywords.

[0027] The matching search engine can use the corresponding text in the webpage interacting with the interactive sentence as a search sentence segment, such as the text between html as a search sentence segment.

[0028] S104: Calculate the number of first keywords in the search sentence segments respectively as the number of first keywords; corresponding to m search sentence segments, obtain m*n numbers of first keywords.

[0029] Among them, one keyword corresponds to one keyword quantity.

[0030] S105: Based on the n first keywords, the m search sentence segments and the corresponding number of first keywords, a pre-match is performed to obtain a plurality of matching sentence segments; the matching sentence segments represent sentence segments that match the answer interaction sentences in the search engine.

[0031] In this embodiment, the matching search engine includes a network search engine and a local search engine.

[0032] S106: Based on multiple matching sentence segments, determine the similarity between the matching sentence segments, and obtain a set of dissimilar sentence segments and multiple sets of similar sentence segments; a similar sentence segment set includes multiple similar sentence segments; the dissimilar sentence segment set includes matching sentence segments that are dissimilar in pairs; the similar sentence segments represent similar matching sentence segments.

[0033] S107: Constructing a first search engine including an index structure based on the set of dissimilar sentence segments and a plurality of sets of similar sentence segments.

[0034] Optionally, the pre-matching is performed based on the n first keywords, the m search statement segments and the corresponding number of first keywords to obtain multiple matching statement segments, including: Randomly obtain a first keyword as the third keyword; The first keyword except the third keyword is used as the second keyword to obtain a second keyword set; the number of elements in the second keyword set is n-1.

[0035] Based on the n first keywords, the m second keyword sets and the corresponding number of first keywords, a graph structure is constructed to obtain a keyword association graph.

[0036] The keyword connectivity graph represents the degree of association between keywords in the search sentence segment and the interactive sentence.

[0037] m search sentence segments correspond to m keyword association graphs.

[0038] Based on the m keyword association graphs and the m search sentence segments, a plurality of matching sentence segments are obtained; the matching sentence segments represent the search sentence segments associated with the interactive sentence.

[0039] Optionally, constructing a graph structure based on the n first keywords, the m second keyword sets and the corresponding number of first keywords to obtain a keyword association graph includes: An association relationship is established between the first keyword and a second keyword in the corresponding second keyword set to obtain a keyword pair; the keyword pair includes a first keyword and a second keyword.

[0040] In this embodiment, the association relationship among the keyword pairs is established by storing the keyword pairs as a set.

[0041] One first keyword corresponds to n-1 keyword pairs; n first keywords correspond to n*(n-1) keyword pairs.

[0042] The two first keyword quantities corresponding to the keyword pair are added together to obtain the keyword pair value.

[0043] n*(n-1) keyword pairs correspond to n*(n-1) keyword pair values.

[0044] Two keyword pairs in which the first keyword and the second keyword are the same are detected, and the two keyword pairs are associated as the same keyword pairs. The keyword pair values ​​corresponding to the same keyword pair are averaged to obtain the average keyword value.

[0045] n*(n-1) keyword pairs correspond to n*(n-1) / 2 average keyword values.

[0046] Based on the n first keywords and the corresponding average keyword values, a keyword association graph is obtained.

[0047] Optionally, obtaining a keyword association graph based on the n first keywords and the corresponding average keyword values ​​includes: Use the n first keywords as rows and columns to construct a connected graph.

[0048] The number of rows and columns of the connected graph is the same, and the first keywords corresponding to the rows and columns are the same.

[0049] According to the first keyword and the second keyword corresponding to the keyword pair, the average keyword value is filled into the corresponding position in the connected graph to obtain a keyword association graph.

[0050] Among them, by constructing a connected graph, it is represented that the graph structure is constructed by taking n first keywords as nodes and the average keyword value as an edge.

[0051] Through the above method, a graph is used to establish the association relationship between the n first keywords in the search sentence segment.

[0052] Optionally, the obtaining of multiple matching sentence segments based on the m keyword association graphs and the m search sentence segments includes: The keyword association graph is input into a first matching network to obtain a first matching degree feature.

[0053] Wherein, the first matching network is a convolutional neural network (CNN).

[0054] The first matching degree feature is input into the first classification network to obtain a first matching degree value; and m first matching degree values ​​are obtained corresponding to the m keyword association graphs.

[0055] Among them, in this embodiment, the first classification network is a softmax function.

[0056] The first matching degree value ranges from 0 to 1.

[0057] The keyword association graph whose first matching degree value is greater than the matching threshold is used as the matching association graph.

[0058] In this embodiment, the matching threshold is 0.85.

[0059] The search sentence segment corresponding to the matching association graph is taken as the matching sentence segment.

[0060] Optionally, the first matching network includes a first convolution kernel, a second convolution kernel and a third convolution kernel.

[0061] The first convolution kernel is a 2*2 two-dimensional convolution kernel; the second convolution kernel is an n*2 two-dimensional convolution kernel; and the third convolution kernel is a 2*n two-dimensional convolution kernel.

[0062] In the direction of rows and columns of the keyword association graph, the first convolution kernel is moved with a step size of 1 for convolution to obtain the first feature graph.

[0063] In the row direction of the keyword association graph, the second convolution kernel is moved with a step size of 1 for convolution to obtain a second feature vector.

[0064] In the column direction of the keyword association graph, the third convolution kernel is moved with a step size of 1 for convolution to obtain a third eigenvector.

[0065] The first feature map, the second feature vector and the third feature vector are input into a first neural network to obtain a first matching degree feature.

[0066] Among them, the first neural network is a fully connected neural network (Fully Connected Neural Network, referred to as FCNN), which can stretch the first feature map into a one-dimensional vector for convolution.

[0067] Through the above method, convolution kernels with different structures are used, and three convolution modes are adopted. Relying on the symmetric characteristics of the keyword association graph, the association characteristics of the keyword association graph can be obtained more accurately.

[0068] m search statement segments correspond to m second keyword sets.

[0069] Optionally, the method of judging the similarity between the matching sentence segments based on the multiple matching sentence segments to obtain a set of dissimilar sentence segments and a set of similar sentence segments includes: In the matching sentence segment, the position of the first keyword is detected to obtain a first key point sequence number; the first key point sequence number represents the sequence number of the text in the matching sentence segment sorted according to the word order.

[0070] Among them, for example, the interactive sentence "What are the common loss functions in deep learning?" matches the matching sentence segment "The loss function in deep learning is used to measure the gap between the model prediction result and the actual label. Common ones are: Mean square error: suitable for regression tasks. Cross entropy: often used in classification tasks. Binary cross entropy: dedicated to binary classification problems.", because the keywords stored in the database include "mean square error", "cross entropy", and "binary cross entropy". The first key point serial numbers are 1 for "mean square error", 2 for "cross entropy", and 3 for "binary cross entropy".

[0071] According to the first key point sequence number, the first keywords corresponding to the matching sentence segments are sorted to obtain a first detection sequence.

[0072] Among them, in this embodiment, according to the first key point sequence number, the first detection sequence is constructed as <"mean square error", "cross entropy", "binary cross entropy">.

[0073] The first detection sequences are input into the discriminant network to judge the similarity and obtain a set of similar sentence segments.

[0074] The discriminant network is a discriminator.

[0075] The matching sentence segments other than the matching sentence segments in the similar sentence segment set are used to construct a dissimilar sentence segment set.

[0076] The non-similar sentence segment set refers to a set including multiple matching sentence segments other than the matching sentence segments in the similar sentence segment set.

[0077] Optionally, constructing a first search engine including an index structure based on the set of dissimilar sentence segments and a plurality of sets of similar sentence segments includes: The index structure includes a first index structure and a second index structure; Taking a matching sentence segment in the similar sentence segment set as the first data to be answered; Putting the first data to be answered into a first index structure; Using the matching sentence segments in the set of dissimilar sentence segments as the second data to be answered; Putting the second data to be answered into a second index structure; Based on the first index structure and the second index structure, a first search engine including an index structure is constructed.

[0078] Optionally, constructing a first search engine including an index structure based on the first index structure and the second index structure includes: The first search engine outputs the first data to be answered in the first index structure; If the first search engine obtains a second interactive statement; the second interactive statement represents a supplement to the interactive statement; The first search engine will be used as a matching search engine in the second index structure to detect the second interactive sentence and obtain a second similar sentence segment set; the second similar sentence segment set represents similar matching sentence segments corresponding to the second interactive sentence.

[0079] The detection method of the second interactive sentence is the same as the detection method of the interactive sentence.

[0080] Example 2 Based on the above-mentioned human-computer dialogue method for database index management based on a large model, an embodiment of the present invention also provides a human-computer dialogue system for database index management based on a large model, the system including an acquisition module, a keyword detection module, a search module, a quantity calculation module, a pre-matching module, a similarity judgment module and a search engine construction module.

[0081] An acquisition module is used to acquire interaction statements; the interaction statements are statements that users interact with during human-computer interaction; A keyword detection module, used for inputting the interactive sentence into the keyword neural network, detecting the keyword, and obtaining n first keywords; A search module, used to detect a sentence segment containing the first keyword by matching a search engine, and obtain m search sentence segments; A quantity calculation module is used to calculate the number of first keywords in the search statement segments as the number of first keywords; m search statement segments correspond to obtaining m*n numbers of first keywords; A pre-matching module performs pre-matching based on the n first keywords, the m search sentence segments and the corresponding number of first keywords to obtain a plurality of matching sentence segments; the matching sentence segments represent sentence segments that match the answering interactive sentence in the search engine; A similarity judgment module is used to judge the similarity between the matching sentence segments based on multiple matching sentence segments, and obtain a set of dissimilar sentence segments and multiple sets of similar sentence segments; a similar sentence segment set includes multiple similar sentence segments; the dissimilar sentence segment set includes matching sentence segments that are not similar to each other; the similar sentence segments represent similar matching sentence segments; The search engine construction module is used to construct a first search engine including an index structure based on the non-similar sentence segment set and a plurality of similar sentence segment sets.

[0082] Similarly, it should be understood that in order to streamline the present disclosure and aid in understanding one or more of the various inventive aspects, in the above description of exemplary embodiments of the present invention, various features of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, this disclosed method should not be interpreted as reflecting the intention that the claimed invention requires more features than those explicitly recited in each claim. More specifically, as reflected in the claims below, inventive aspects lie in less than all the features of the individual embodiments disclosed above. Therefore, the claims that follow the specific embodiment are hereby expressly incorporated into the specific embodiment, with each claim itself serving as a separate embodiment of the present invention.

Claims

1. A human-computer dialogue method for database index management based on a large model, characterized in that: include: Obtaining an interactive statement; the interactive statement is a statement for a user to interact during human-computer interaction; Input the interactive sentences into the keyword neural network, detect the keywords, and obtain n first keywords; By matching the search engine, the sentence segments containing the first keyword are detected to obtain m search sentence segments; Calculate the number of first keywords in the search statement segments respectively as the number of first keywords; m search statement segments correspond to obtaining m*n first keywords; Based on the n first keywords, the m search sentence segments and the corresponding number of first keywords, a pre-match is performed to obtain a plurality of matching sentence segments; The matching sentence segment represents a sentence segment that matches the answering interactive sentence in the search engine; Based on multiple matching sentence segments, the similarity between the matching sentence segments is determined to obtain a set of dissimilar sentence segments and a set of multiple similar sentence segments; A similar sentence segment set contains multiple similar sentence segments; The dissimilar sentence segment set includes two dissimilar matching sentence segments; The similar sentence segments represent similar matching sentence segments; Based on the dissimilar sentence segment set and a plurality of similar sentence segment sets, a first search engine including an index structure is constructed.

2. According to the human-computer dialogue method for database index management based on a large model according to claim 1, it is characterized in that: The method performs a pre-match based on the n first keywords, the m search sentence segments and the corresponding number of first keywords to obtain a plurality of matching sentence segments, including: Randomly obtain a first keyword as the third keyword; The first keyword except the third keyword is used as the second keyword to obtain a second keyword set; the number of elements in the second keyword set is n-1; Based on the n first keywords, the m second keyword sets and the corresponding number of first keywords, a graph structure is constructed to obtain a keyword association graph; m search sentence segments correspond to m keyword association graphs; Based on the m keyword association graphs and the m search sentence segments, a plurality of matching sentence segments are obtained; the matching sentence segments represent the search sentence segments associated with the interactive sentence.

3. The human-computer dialogue method for database index management based on a large model according to claim 2 is characterized in that: The method of constructing a graph structure based on the n first keywords, the m second keyword sets and the corresponding number of first keywords to obtain a keyword association graph includes: Establishing an association relationship between the first keyword and a second keyword in the corresponding second keyword set to obtain a keyword pair; the keyword pair includes a first keyword and a second keyword; One first keyword corresponds to n-1 keyword pairs; n first keywords correspond to n*(n-1) keyword pairs; Add the two first keyword quantities corresponding to the keyword pair to obtain the keyword pair value; n*(n-1) keyword pairs correspond to n*(n-1) keyword pair values; Detecting two keyword pairs in which the first keyword and the second keyword are identical, and associating the two keyword pairs as identical keyword pairs; Average the keyword pair values ​​corresponding to the same keyword pair to obtain the average keyword value; For n*(n-1) keyword pairs, n*(n-1) / 2 average keyword values ​​are obtained; Based on the n first keywords and the corresponding average keyword values, a keyword association graph is obtained.

4. The human-computer dialogue method for database index management based on a large model according to claim 2 is characterized in that: The step of obtaining a keyword association graph based on the n first keywords and the corresponding average keyword values ​​includes: Use the n first keywords as rows and columns to construct a connected graph; The number of rows and columns of the connected graph is the same, and the first keywords corresponding to the rows and columns are the same; According to the first keyword and the second keyword corresponding to the keyword pair, the average keyword value is filled into the corresponding position in the connected graph to obtain a keyword association graph.

5. The human-computer dialogue method for database index management based on a large model according to claim 2 is characterized in that: The method of obtaining multiple matching sentence segments based on the m keyword association graphs and the m search sentence segments includes: Inputting the keyword association graph into a first matching network to obtain a first matching degree feature; Inputting the first matching degree feature into the first classification network to obtain a first matching degree value; and obtaining m first matching degree values ​​corresponding to the m keyword association graphs; The keyword association graph whose first matching degree value is greater than the matching threshold is used as the matching association graph; The search sentence segment corresponding to the matching association graph is taken as the matching sentence segment.

6. The human-computer dialogue method for database index management based on a large model according to claim 5 is characterized in that: The first matching network includes a first convolution kernel, a second convolution kernel and a third convolution kernel; The first convolution kernel is a 2*2 two-dimensional convolution kernel; the second convolution kernel is an n*2 two-dimensional convolution kernel; the third convolution kernel is a 2*n two-dimensional convolution kernel; In the direction of rows and columns of the keyword association graph, the first convolution kernel is moved with a step size of 1 for convolution to obtain the first feature graph; In the row direction of the keyword association graph, the second convolution kernel is moved with a step size of 1 to perform convolution to obtain a second eigenvector; In the column direction of the keyword association graph, the third convolution kernel is moved with a step size of 1 to perform convolution to obtain a third eigenvector; The first feature map, the second feature vector and the third feature vector are input into a first neural network to obtain a first matching degree feature.

7. The human-computer dialogue method for database index management based on a large model according to claim 1 is characterized in that: The method of judging the similarity between the matching sentence segments based on the multiple matching sentence segments to obtain a set of dissimilar sentence segments and a set of similar sentence segments includes: In the matching sentence segment, the position of the keyword is detected to obtain a first key point sequence number; the first key point sequence number represents the sequence number of the keyword in the matching sentence segment sorted according to the word order; According to the first key point sequence number, the keywords corresponding to the matching sentence segments are sorted to obtain a first detection sequence; Inputting the first detection sequences in pairs into the discriminant network to judge the similarity, and obtaining a plurality of first similar sentence segment sets; the first similar sentence segment sets include two similar matching sentence segments; Obtaining an intersection of multiple first similar sentence segment sets to obtain a similar sentence segment set; The matching sentence segments other than the matching sentence segments in the similar sentence segment set are used to construct a dissimilar sentence segment set.

8. The human-computer dialogue method for database index management based on a large model according to claim 1 is characterized in that: The step of constructing a first search engine including an index structure based on the set of dissimilar sentence segments and a plurality of sets of similar sentence segments includes: The index structure includes a first index structure and a second index structure; Taking a matching sentence segment in the similar sentence segment set as the first data to be answered; Putting the first data to be answered into a first index structure; Using the matching sentence segments in the set of dissimilar sentence segments as the second data to be answered; Putting the second data to be answered into a second index structure; Based on the first index structure and the second index structure, a first search engine including an index structure is constructed.

9. The human-computer dialogue method for database index management based on a large model according to claim 8, characterized in that: The step of constructing a first search engine including an index structure based on the first index structure and the second index structure includes: The first search engine outputs the first data to be answered in the first index structure; If the first search engine obtains a second interactive statement; the second interactive statement represents a supplement to the interactive statement; The first search engine will be used as a matching search engine in the second index structure to detect the second interactive sentence and obtain a second similar sentence segment set; the second similar sentence segment set represents similar matching sentence segments corresponding to the second interactive sentence.

10. A human-computer dialogue system for database index management based on a large model, characterized in that: include: An acquisition module is used to acquire interaction statements; the interaction statements are statements that users interact with during human-computer interaction; A keyword detection module, used for inputting the interactive sentence into the keyword neural network, detecting the keyword, and obtaining n first keywords; A search module, used to detect a sentence segment containing the first keyword by matching a search engine, and obtain m search sentence segments; A quantity calculation module, used to calculate the quantity of the first keywords in the search sentence segments as the first keyword quantity; m search statement segments correspond to obtaining m*n first keywords; A pre-matching module performs pre-matching based on the n first keywords, the m search sentence segments and the corresponding number of first keywords to obtain a plurality of matching sentence segments; The matching sentence segment represents a sentence segment that matches the answering interactive sentence in the search engine; A similarity judgment module is used to judge the similarity between the matching sentence segments based on the multiple matching sentence segments, and obtain a set of dissimilar sentence segments and a set of multiple similar sentence segments; A similar sentence segment set contains multiple similar sentence segments; The dissimilar sentence segment set includes two dissimilar matching sentence segments; The similar sentence segments represent similar matching sentence segments; The search engine construction module is used to construct a first search engine including an index structure based on the non-similar sentence segment set and a plurality of similar sentence segment sets.

Images