Conversation service method and system for user intention recognition in social interaction

Abstract

The invention discloses a dialogue service method and system for user intention recognition in social interaction, and relates to the technical field of intention recognition, word segmentation processing is performed on a dialogue text of a target user, an intention appeal of the target user is determined according to a word vector, and the intention appeal of the target user is inferred according to the word vector of the dialogue text of the target user. Obtaining the intention appeal of the target user and the weight corresponding to each intention appeal, and constructing an intention appeal set; obtaining the intention of the target user based on the intention appeal set of the target user; performing clustering analysis on the word vector of the target user, determining a cluster to which the word vector of the target user belongs, obtaining the weight of the intention appeal from the cluster, and removing the intention of the target user according to the intention appeal directly obtained from the text and the intention appeal obtained through inference; the accuracy of identifying the user intention is improved, the limitation of identifying the user intention is reduced, and a reference is provided for displaying recommended information.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intention recognition, in particular to a conversation service method and system for user intention recognition in social interaction. BACKGROUND

[0002] In social interaction, recognizing user intention is of great significance, which can improve interaction efficiency and enable users to quickly achieve goals such as obtaining information and completing tasks. Social platforms use natural language processing technology to understand the text content of users in social interaction, and can recognize the interest intention expressed by the user, and recommend content to the user according to the interest intention of the user. Intention recognition makes social interaction more efficient and can improve the social experience of users. However, when the intention information contained in the conversation text of the user is less, the intention category of the user is not clear, which affects the efficiency of intention recognition. Therefore, how to quickly and accurately recognize the intention of the user and show the recommended content to the user has become a problem to be solved. SUMMARY

[0003] The present application aims to provide a conversation service method and system for user intention recognition in social interaction to solve the problems in the prior art.

[0004] To achieve the above-mentioned purpose, the present application provides the following technical scheme: a conversation service system for user intention recognition in social interaction, comprising an input module, a text processing module, an analysis module, an output module and a data storage module; the input module is used to obtain the conversation text input by a target user; the data storage module is used to store the conversation text input by a historical user and the intention appeal information of the historical user; the text processing module is used to perform word segmentation processing on the conversation text input by the target user and calculate the word vector of each word segmentation; the analysis module determines the intention appeal set and semantic information of the target user based on the word vector of the conversation text input by the target user, and obtains the intention of the target user based on the intention appeal set and the semantic information; the output module shows the recommended information to the target user based on the intention of the target user.

[0005] Specifically, the text processing module further comprises a preprocessing unit, a word segmentation unit and a word vector calculation unit, the preprocessing unit is used to preprocess the conversation text input by the target user and correct errors in spelling and grammar; the word segmentation unit is used to perform word segmentation processing on the conversation text input by the target user to obtain a plurality of word segments; and the word vector calculation unit calculates the word vector according to the word segments.

[0006] Specifically, the analysis module further comprises an intention appeal analysis unit, a semantic analysis unit and an intention determination unit; the intention appeal analysis unit determines an intention appeal set of the target user based on the word vector of the dialogue text of the target user; the semantic analysis unit determines semantic information based on the word vector of the dialogue text of the target user to obtain the intention appeal contained in the dialogue text; and the intention determination unit combines the intention appeal set and the semantic information set of the target user to obtain the intention of the target user.

[0007] Specifically, the intention appeal analysis unit constructs an intention appeal vector sequence and a weight sequence according to the intention appeals in the intention appeal set, obtains a weighted intention appeal vector by weighted summation of the intention appeal vector sequence and the weight sequence, inputs the weighted intention appeal vector into an intention recognition model to obtain a probability distribution of the intention category, and determines the intention of the target user according to the probability distribution of the intention category.

[0008] To achieve the above object, the present application provides the following technical scheme: a dialogue service method for user intention recognition in social interaction, comprising the following steps:

[0009] The dialogue text of the target user is obtained, the dialogue text of the target user is processed by word segmentation, the obtained word segmentation is represented by a word vector, the intention appeal of the target user is determined according to the word vector, the intention appeal of the target user is inferred according to the word vector of the dialogue text of the target user, the intention appeal of the target user and the weight corresponding to each intention appeal are obtained, and an intention appeal set is constructed; the intention of the target user is obtained based on the intention appeal set of the target user.

[0010] Specifically, determining the intention appeal of the target user according to the word vector further comprises the following steps:

[0011] S11, it is judged whether the dialogue text of the target user contains an intention appeal, if the dialogue text of the target user contains an intention appeal, the intention appeal contained in the dialogue text of the target user is added to the intention appeal set, and the weight of the intention appeal contained in the dialogue text is set to 1; if the dialogue text of the target user does not contain an intention appeal, the intention appeal is not added to the intention appeal set;

[0012] S12, the intention appeal and the dialogue text of the historical user are obtained, the dialogue text of the historical user is processed by word segmentation to obtain a plurality of word segmentation; a plurality of word vectors are calculated according to the obtained word segmentation, and the dialogue text of the historical user is represented by the word vector;

[0013] S13, the word vector is taken as input, the DBSCAN algorithm is used to cluster the word vector of the historical user to obtain n clustering clusters and m outliers, for the jth data point in the ith clustering cluster, the number num of data points belonging to the ith clustering cluster in the neighborhood radius range of the jth data point is obtained j, according to the different values of j, the obtained num j Minimum value num min ;

[0014] S14, the target user dialogue text word vector is obtained in the manner of step S12, the cluster cluster to which the target user dialogue text word vector belongs is determined, the target user intention appeal and the corresponding weight are determined according to the cluster cluster; The determined target user intention appeal and the corresponding weight are added to the intention appeal set.

[0015] Specifically, the neighborhood radius of the DBSCAN algorithm is determined by the following steps:

[0016] According to the dimension of the word vector, the k value of the k nearest neighbor algorithm is determined; the Euclidean distance between each word vector and the kth nearest other word vector is calculated, and the obtained Euclidean distance is expressed in a plane rectangular coordinate system in descending order, the vertical coordinate is the Euclidean distance, and the horizontal coordinate is a positive integer; When the horizontal coordinate is 1, the corresponding Euclidean distance is the maximum value of the Euclidean distance from all word vectors to the kth nearest word vector, and when the horizontal coordinate is the maximum value, the corresponding Euclidean distance is the minimum value of the Euclidean distance from all word vectors to the kth nearest word vector.

[0017] The horizontal coordinate is denoted as x, and the vertical coordinate y is denoted as y. Two points in the rectangular coordinate system are selected, and the first-order difference is calculated according to the selected two points, and all first-order differences are obtained. For two first-order differences involving the same point in the rectangular coordinate system, the second-order difference is calculated, the maximum value is obtained from all second-order differences, and the inflection point is determined according to the position of the maximum value. The vertical coordinate of the inflection point is the neighborhood radius.

[0018] Specifically, determining the cluster cluster to which the target user dialogue text word vector belongs further includes the following steps:

[0019] S21, obtaining the number n of historical user word vectors belonging to the i-th cluster cluster within the neighborhood radius range of the target user dialogue text word vector i , if n i is not less than num min , the target user dialogue text word vector belongs to the i-th cluster cluster, otherwise the target user dialogue text word vector does not belong to the i-th cluster cluster; All cluster clusters obtained in step S13 are analyzed to determine all cluster clusters to which the target user dialogue text word vector belongs; If the target user dialogue text word vector does not belong to any existing cluster cluster, it is judged whether the target user dialogue text word vector can form a new cluster cluster: within the neighborhood radius range of the target user dialogue text word vector, if there are at least M outliers, M represents the minimum number of points of the DBSCAN algorithm, then the target user dialogue text word vector can form a new cluster cluster, otherwise the target user dialogue text word vector cannot form a new cluster cluster;

[0020] S22, if the target user dialogue text word vector only belongs to one existing or new cluster, set the weight of the cluster to which the target user dialogue text word vector belongs to one; if the target user dialogue text word vector belongs to a existing or new cluster, a is a positive integer greater than one, then record the a existing or new cluster as a candidate cluster and determine the weight of each candidate cluster: obtain the number of word vectors belonging to the u-th candidate cluster within the neighborhood radius range of the target user dialogue text word vector r u , according to the weight of the u-th candidate cluster Q u , Q u = (r u / adis u ) / Σ u (r u / adis u ), wherein adis u represents the average value of the Euclidean distance between the word vector of the u-th candidate cluster and the target user dialogue text word vector;

[0021] S23, if the target user dialogue text word vector does not belong to any existing or new cluster, obtain the context information of the target user dialogue text, and perform steps S21 and S22 on the context.

[0022] Specifically, determining the target user intent appeal and the corresponding weight according to the cluster to which it belongs further comprises the following steps:

[0023] Obtain the cluster to which the target user dialogue text word vector belongs and the corresponding weight, and in the b-th cluster, determine the d-th intent appeal corresponding to the word vector in the b-th cluster, and obtain the weight y(b,d) of the d-th intent appeal according to the frequency of the d-th intent appeal, y(b,d)=f(b,d)×Q b , wherein f(b,d) represents the frequency of the d-th intent appeal in the b-th cluster, Q b represents the weight of the b-th cluster, and obtain the maximum weight y(d) of the d-th intent appeal according to y(b,d), y(d)=Σ b y(b,d); add all intent appeals and corresponding maximum weights to the intent appeal set.

[0024] Specifically, obtaining the target user's intent based on the target user's intent appeal set comprises the following steps:

[0025] Obtaining the intention appeal and the dialogue text of the historical user, performing word segmentation on the dialogue text of the historical user to obtain a plurality of segmented words, calculating the word vectors according to the segmented words, generating a training set according to the word vectors and the intention appeal, and generating an intention recognition model according to the training set;

[0026] According to the intention appeal in the intention appeal set, an intention appeal vector sequence and a weight sequence are constructed, a weighted intention appeal vector is obtained by weighted summation on the intention appeal vector sequence and the weight sequence, the weighted intention appeal vector is input into the intention recognition model, and the probability distribution of the intention category is obtained; and the intention of the target user is determined according to the probability distribution of the intention category.

[0027] Compared with the prior art, the beneficial effects of the present application are as follows: the intention appeal information of the target user is supplemented according to the word vector of the target user; the word vector of the target user is subjected to cluster analysis to determine the cluster cluster to which the word vector of the target user belongs, the weight of the intention appeal is obtained from the cluster cluster, and the intention of the target user is removed according to the intention appeal directly obtained from the text and the intention appeal obtained by inference; the accuracy of recognizing the user intention is improved, the limitation of recognizing the user intention is reduced, and reference is provided for displaying the recommended information. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 The figure is a structural schematic diagram of a dialogue service system for user intention recognition in social interaction according to the present application.

[0029] Figure 2 The figure is a flowchart of a dialogue service method for user intention recognition in social interaction according to the present application. DETAILED DESCRIPTION

[0030] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0031] Embodiment: as Figure 1As shown, the present application provides a dialogue service system for user intention recognition in social interaction, comprising an input module, a text processing module, an analysis module, an output module and a data storage module; the input module is used to obtain the dialogue text input by the target user; the data storage module is used to store the dialogue text input by the historical user and the intention appeal information of the historical user; the text processing module is used to perform word segmentation processing on the dialogue text input by the target user, and calculate the word vector of each word segmentation; the analysis module determines the intention appeal set and the semantic information of the target user based on the word vector of the dialogue text input by the target user, and obtains the intention of the target user based on the intention appeal set and the semantic information; the output module shows the recommended information to the target user based on the intention of the target user.

[0032] The text processing module further comprises a preprocessing unit, a word segmentation unit and a word vector calculation unit, the preprocessing unit is used to preprocess the dialogue text input by the target user, and correct the wrong words and grammatical errors; the word segmentation unit is used to perform word segmentation processing on the dialogue text input by the target user, and obtain a plurality of word segments; the word vector calculation unit calculates the word vector according to the word segmentation. The analysis module further comprises an intention appeal analysis unit, a semantic analysis unit and an intention determination unit; the intention appeal analysis unit determines the intention appeal set of the target user based on the word vector of the dialogue text of the target user; the semantic analysis unit determines the semantic information based on the word vector of the dialogue text of the target user, and obtains the intention appeal contained in the dialogue text; the intention determination unit combines the intention appeal set and the semantic information set of the target user, and obtains the intention of the target user. The intention appeal analysis unit constructs an intention appeal vector sequence and a weight sequence according to the intention appeal in the intention appeal set, obtains the probability distribution of the intention category by weighting and summing the intention appeal vector sequence and the weight sequence, inputs the weighted intention appeal vector into the intention recognition model, and determines the intention of the target user according to the probability distribution of the intention category.

[0033] In another embodiment of the present application, the present application provides a dialogue service method for user intention recognition in social interaction, comprising the following steps:

[0034] Obtaining the dialogue text of the target user; performing word segmentation processing on the dialogue text of the target user, representing the obtained word segmentation by a word vector, determining the intention appeal of the target user according to the word vector, inferring the intention appeal of the target user according to the word vector of the dialogue text of the target user, obtaining the intention appeal of the target user and the weight corresponding to each intention appeal, and constructing an intention appeal set; obtaining the intention of the target user based on the intention appeal set of the target user.

[0035] Determining the intention appeal of the target user according to the word vector further comprises the following steps:

[0036] S11, determine whether the target user's dialogue text contains an intent appeal, if the intent appeal is contained in the target user's dialogue text, add the intent appeal contained in the target user's dialogue text to the intent appeal set, and set the weight of the intent appeal contained in the dialogue text to 1; if the intent appeal is not contained, do not add it to the intent appeal set;

[0037] S12, obtain the intent appeal and dialogue text of the historical user, perform word segmentation processing on the dialogue text of the historical user to obtain a plurality of segmented words; calculate a plurality of word vectors according to the obtained segmented words, and represent the dialogue text of the historical user by the word vectors;

[0038] S13, take the word vector as input, and cluster the word vector of the historical user by DBSCAN algorithm to obtain n clustering clusters and m outliers, for the jth data point in the ith clustering cluster, obtain the number num j of data points belonging to the ith clustering cluster within the neighborhood radius of the jth data point according to different values of j j ; min Obtain the minimum value num

[0039] S14, obtain the target user dialogue text word vector in the manner of step S12, determine the clustering cluster to which the target user dialogue text word vector belongs, determine the target user intent appeal and the corresponding weight according to the belonging clustering cluster; add the determined target user intent appeal and the corresponding weight to the intent appeal set.

[0040] For the intent appeal contained in the dialogue text of the target user, the weight is 1; however, the intent of the target user may not be clear, or the dialogue text of the target user does not contain the intent, so it is necessary to mine the target user; determine the intent appeal of the historical user with similar word vectors of the target user through DBSCAN algorithm, and infer the intent appeal of the target user.

[0041] The neighborhood radius of the DBSCAN algorithm is determined by the following steps:

[0042] According to the dimension of the word vector, determine the k value of the k nearest neighbor algorithm; calculate the Euclidean distance between each word vector and the kth nearest other word vector, and represent the obtained Euclidean distance in a plane rectangular coordinate system in descending order, with the vertical coordinate as the Euclidean distance and the horizontal coordinate as a positive integer; when the horizontal coordinate is 1, the corresponding Euclidean distance is the maximum value of the Euclidean distance from all word vectors to the kth nearest word vector, and when the horizontal coordinate is the maximum value, the corresponding Euclidean distance is the minimum value of the Euclidean distance from all word vectors to the kth nearest word vector;

[0043] The horizontal coordinate is marked as x, the vertical coordinate y is marked as y, two points with a difference of 1 between the horizontal coordinates in the rectangular coordinate system are selected, a first-order difference is calculated according to the selected two points, and all first-order differences are obtained; for two first-order differences involving the same point in the rectangular coordinate system, a second-order difference is calculated, the maximum value is obtained from all second-order differences, and the inflection point is determined according to the position of the maximum value, and the vertical coordinate of the inflection point is the neighborhood radius.

[0044] The value of k can be set to 2xword vector dimension-1, according to the value of k, a k-distance image can be drawn, and the position of the inflection point is determined from the image, and the Euclidean distance corresponding to the position of the inflection point is also the neighborhood radius. The maximum value of the second-order difference is obtained by performing twice difference, and the position of the inflection point is determined according to the maximum value of the second-order difference, and the neighborhood radius is obtained.

[0045] If there are 50 word vectors, each word vector can obtain the Euclidean distance between the other 49 word vectors, for a word vector, the 49 Euclidean distances obtained are arranged in ascending order, and the Euclidean distance with the serial number k in the arrangement is obtained according to the arrangement, which is the Euclidean distance between the word vector and the other word vector closest to the kth, that is, the k-distance. The 50 k-distances obtained are arranged in descending order in the rectangular coordinate system, and each two adjacent k-distances can obtain a first-order difference, and each two first-order differences can obtain a second-order difference, for example, the first and second k-distances and the second and third k-distances in the k-distance image each obtain a first-order difference, and the two first-order differences contain the same second k-distance, according to the two first-order differences obtained, a second-order difference is obtained, if the second-order difference is the maximum value, the position of the inflection point is the position of the second k-distance, and the second k-distance is the neighborhood radius.

[0046] Determining the cluster cluster to which the target user dialogue text word vector belongs further includes the following steps:

[0047] S21, obtaining the number n of historical user word vectors belonging to the i-th cluster cluster within the neighborhood radius range of the target user dialogue text word vector i , if n i is not less than num min, the target user dialogue text word vector belongs to the i-th clustering cluster, otherwise the target user dialogue text word vector does not belong to the i-th clustering cluster; all clustering clusters obtained in step S13 are analyzed to determine all clustering clusters to which the target user dialogue text word vector belongs; if the target user dialogue text word vector does not belong to any existing clustering cluster, it is determined whether the target user dialogue text word vector can form a new clustering cluster; if there are at least M outliers within the neighborhood radius range of the target user dialogue text word vector, M representing the minimum number of points of the DBSCAN algorithm, the target user dialogue text word vector can form a new clustering cluster, otherwise the target user dialogue text word vector cannot form a new clustering cluster;

[0048] S22, if the target user dialogue text word vector only belongs to one existing or new clustering cluster, the weight of the clustering cluster to which the target user dialogue text word vector belongs is set to one; if the target user dialogue text word vector belongs to a existing or new clustering cluster, a is a positive integer greater than one, the a existing or new clustering clusters are recorded as candidate clustering clusters and the weight of each candidate clustering cluster is determined; the number of word vectors belonging to the u-th candidate clustering cluster within the neighborhood radius range of the target user dialogue text word vector is obtained u , the weight of the u-th candidate clustering cluster is obtained according to u , Q u = (r u / adis u ) / ∑ u (r u / adis u ), wherein adis u represents the average value of the Euclidean distance between the word vector of the u-th candidate clustering cluster and the target user dialogue text word vector;

[0049] S23, if the target user dialogue text word vector does not belong to any existing or new clustering cluster, the context information of the target user dialogue text is obtained, and steps S21 and S22 are performed on the context.

[0050] Optionally, set M as k+1; in the case that the target user dialogue text word vector does not belong to any cluster, according to the principle of DBSCAN clustering algorithm, if there are not less than M data points in the neighborhood radius range of a data point, the data point is a core point and can form a cluster, so the intention appeal of the target user can be analyzed according to the intention appeal of the historical user contained in the formed cluster; if the target user dialogue text word vector belongs to an existing cluster, but the cluster is not unique, then the weights of multiple clusters are analyzed according to the word vectors in the neighborhood radius range of the target user dialogue text word vector; in the case that the properties of other word vectors in the neighborhood radius range of the target user dialogue text word vector are similar to those of the target user dialogue text word vector, the weights of multiple clusters to which the target user dialogue text word vector belongs are determined according to the clusters to which the word vectors with similar properties belong; the farther the Euclidean distance, the lower the property similarity, so it is inversely related to the average Euclidean distance and positively related to the number of word vectors in the cluster.

[0051] According to the clusters to which the target user belongs, the target user's intention appeal and the corresponding weight are determined, which further includes the following steps:

[0052] The clusters to which the target user dialogue text word vector belongs and the corresponding weights are obtained, in the bth cluster, the dth intention appeal corresponding to the word vector in the bth cluster is determined, and the weight y(b,d) of the dth intention appeal is obtained according to the frequency of the dth intention appeal, y(b,d)=f(b,d)×Q b , wherein f(b,d) represents the frequency of the dth intention appeal in the bth cluster, Q b represents the weight of the bth cluster, and the maximum weight y(d) of the dth intention appeal is obtained according to y(b,d), y(d)=∑ b y(b,d); all intention appeals and the corresponding maximum weights are added to the intention appeal set.

[0053] In each cluster, the intention appeal of the historical user can be obtained, the weight of the intention appeal is obtained according to the frequency of the intention appeal and the weight of the cluster, and the maximum weight of the intention appeal is obtained by adding the weights of the same intention appeal in different clusters.

[0054] The intention of the target user is obtained based on the intention appeal set of the target user, which includes the following steps:

[0055] The intention appeal of the historical user and the dialogue text are obtained, the dialogue text of the historical user is segmented to obtain a plurality of segmented words; the word vector is calculated according to the segmented words, the training set is generated according to the word vector and the intention appeal, and the intention recognition model is generated according to the training set;

[0056] The intention appeal vector sequence and the weight sequence are constructed according to the intention appeals in the intention appeal set, the weighted intention appeal vector is obtained by weighted summation on the intention appeal vector sequence and the weight sequence, the weighted intention appeal vector is input into the intention recognition model, and the probability distribution of the intention category is obtained; and the intention of the target user is determined according to the probability distribution of the intention category.

[0057] The intention appeal is also represented in the form of a word vector, the intention appeal in the intention appeal set is directly obtained from the dialogue text, which is an explicit intention appeal, and therefore the weight is one; and the intention appeal obtained by inference has a weight between 0 and 1, the discrete intention appeal is converted into an overall intention appeal, and the intention of the target user is determined according to the overall intention appeal.

[0058] It is apparent for a person skilled in the art that the application is not restricted to the details of the exemplary embodiments described above, but that it can be implemented in other embodiments without deviating from the spirit or essential characteristics of the application. Therefore, the embodiments should be seen as exemplary only and not limiting, the scope of the application being defined by the claims below rather than by the above description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. No reference signs in the claims should be considered as limiting the scope of the claims.

Claims

1. A dialogue service method for recognizing user intent in social interactions, characterized in that, Includes the following steps: Obtain the target user's conversation text; The target user's dialogue text is segmented into words, and the segmented words are represented by word vectors. The target user's intent is determined based on the word vectors. The target user's intent is inferred based on the word vectors of the dialogue text, and the target user's intent and the weight of each intent are obtained, thus constructing a set of intents. The intent of the target user is obtained based on the set of intent requests of the target user.

2. The dialogue service method for user intent recognition in social interaction according to claim 1, characterized in that, Determining the target user's intent based on word vectors also includes the following steps: S11, determine whether the target user's dialogue text contains an intent request. If it contains an intent request, add the intent request contained in the target user's dialogue text to the intent request set and set the weight of the intent request contained in the dialogue text to 1; if it does not contain an intent request, do not add it to the intent request set. S12: Obtain the intent and dialogue text of historical users, perform word segmentation on the dialogue text of historical users to obtain several words; calculate several word vectors based on the obtained word segments, and use word vectors to represent the dialogue text of historical users. S13, using word vectors as input, cluster the word vectors of historical users using the DBSCAN algorithm to obtain n clusters and m outliers. For the j-th data point in the i-th cluster, obtain the number num of data points belonging to the i-th cluster within the neighborhood radius of the j-th data point. j Based on the different values ​​of j, the obtained num j Minimum value num min ; S14. Obtain the target user's dialogue text word vectors in accordance with step S12, determine the cluster to which the target user's dialogue text word vectors belong, determine the target user's intent and corresponding weights based on the cluster to which they belong, and add the determined target user intent and corresponding weights to the intent set.

3. The dialogue service method for user intent recognition in social interaction according to claim 2, characterized in that, The neighborhood radius of the DBSCAN algorithm is determined through the following steps: Based on the dimension of the word vectors, determine the value of k for the k-nearest neighbor algorithm; calculate the Euclidean distance between each word vector and the other word vectors that are k closest to it, and represent the obtained Euclidean distances in descending order on a Cartesian coordinate system, with the ordinate being the Euclidean distance and the abscissa being a positive integer; when the abscissa is 1, the corresponding Euclidean distance is the maximum value of the Euclidean distance from all word vectors to the k closest word vector, and when the abscissa is the maximum value, the corresponding Euclidean distance is the minimum value of the Euclidean distance from all word vectors to the k closest word vector; Let x be the horizontal coordinate and y be the vertical coordinate. Select two points in the rectangular coordinate system whose horizontal coordinates differ by 1. Calculate the first-order difference based on the two selected points to obtain all the first-order differences. For two first-order differences involving the same point in the rectangular coordinate system, calculate the second-order difference. Obtain the maximum value from all the second-order differences. Determine the inflection point based on the position of the maximum value. The vertical coordinate of the inflection point is the neighborhood radius.

4. The dialogue service method for user intent recognition in social interaction according to claim 3, characterized in that, In step S14, determining the cluster to which the target user's dialogue text word vectors belong further includes the following steps: S21, obtain the number n of historical user word vectors belonging to the i-th cluster within the neighborhood radius of the target user's dialogue text word vectors. i If n i Not less than num min If the target user dialogue text word vector belongs to the i-th cluster, then the target user dialogue text word vector does not belong to the i-th cluster. Analyze all the clusters obtained in step S13 to determine all the clusters to which the target user dialogue text word vector belongs. If the target user dialogue text word vector does not belong to any existing cluster, then determine whether the target user dialogue text word vector can form a new cluster: within the neighborhood radius of the target user dialogue text word vector, if there are at least M outliers, where M represents the minimum number of points in the DBSCAN algorithm, then the target user dialogue text word vector can form a new cluster; otherwise, the target user dialogue text word vector cannot form a new cluster. S22, if the target user's dialogue text word vector belongs to only one existing or new cluster, then reset the weight of the cluster to which the target user's dialogue text word vector belongs to one; if the target user's dialogue text word vector belongs to *a* existing or new clusters, where *a* is a positive integer greater than one, then record *a* existing or new clusters as candidate clusters and determine the weight of each candidate cluster: obtain the number *r* of word vectors belonging to the *u*-th candidate cluster within the neighborhood radius of the target user's dialogue text word vector. u Based on the weight Q of the u-th candidate cluster u Q u =(r u / adis u ) / Σ u (r u / adis u ), where adis u This represents the average Euclidean distance between the word vectors of the u-th candidate cluster and the word vectors of the target user's dialogue text; S23, if the word vector of the target user's dialogue text does not belong to any existing or new cluster, then obtain the context information of the target user's dialogue text, and proceed with steps S21 and S22 for the context.

5. The dialogue service method for user intent recognition in social interaction according to claim 4, characterized in that, Determining the target user's intent and corresponding weight based on the cluster they belong to also includes the following steps: Obtain the clusters and corresponding weights of the word vectors in the target user's dialogue text. Within the b-th cluster, determine the d-th intent request corresponding to the word vectors in the b-th cluster. Based on the frequency of the d-th intent request, obtain its weight y(b,d), where y(b,d) = f(b,d) × Q. b In the formula, f(b,d) represents the frequency of the d-th intentional request in the b-th cluster, and Q b This represents the weight of the b-th cluster. The final weight y(d) of the d-th intention is obtained from y(b,d), where y(d) = Σ b y(b,d); Add all intent claims and their corresponding final weights to the intent claim set.

6. The dialogue service method for user intent recognition in social interaction according to claim 5, characterized in that, The process of obtaining the target user's intent based on the target user's intent request set includes the following steps: The process involves acquiring the intent requests and dialogue texts of historical users, performing word segmentation on the dialogue texts to obtain several word segments, calculating word vectors based on the obtained word segments, generating a training set based on the word vectors and intent requests, and generating an intent recognition model based on the training set. Based on the intent requests in the intent request set, an intent request vector sequence and a weight sequence are constructed. The weighted intent request vector is obtained by weighted summation of the intent request vector sequence and the weight sequence. The weighted intent request vector is then input into the intent recognition model to obtain the probability distribution of intent categories. The intent of the target user is determined based on the probability distribution of intent categories.

7. A dialogue service system for recognizing user intent in social interactions, characterized in that, It includes an input module, a text processing module, an analysis module, an output module, and a data storage module; the input module is used to acquire the dialogue text input by the target user; the data storage module is used to store the dialogue text input by historical users and the intent request information of historical users; the text processing module is used to perform word segmentation on the dialogue text input by the target user and calculate the word vector of each segment. The analysis module determines the target user's intent request set and semantic information based on the word vectors of the dialogue text input by the target user, and obtains the target user's intent based on the intent request set and semantic information; The output module displays recommended information to the target user based on the target user's intent.

8. The dialogue service system for user intent recognition in social interaction according to claim 7, characterized in that, The text processing module further includes a preprocessing unit, a word segmentation unit, and a word vector calculation unit. The preprocessing unit is used to preprocess the dialogue text input by the target user and correct typos and grammatical errors. The word segmentation unit is used to segment the dialogue text input by the target user to obtain several words. The word vector calculation unit calculates word vectors based on the word segmentation.

9. The dialogue service system for user intent recognition in social interaction according to claim 7, characterized in that, The analysis module further includes an intent appeal analysis unit, a semantic analysis unit, and an intent determination unit; the intent appeal analysis unit determines the target user's intent appeal set based on the word vectors of the target user's dialogue text; The semantic analysis unit determines semantic information based on the word vectors of the target user's dialogue text to obtain the intent requests contained in the dialogue text; the intent determination unit merges the target user's intent request set and semantic information set to obtain the target user's intent.

10. The dialogue service system for user intent recognition in social interaction according to claim 9, characterized in that, The intent appeal analysis unit constructs an intent appeal vector sequence and a weight sequence based on the intent appeals in the intent appeal set. By weighted summing of the intent appeal vector sequence and the weight sequence, a weighted intent appeal vector is obtained. The weighted intent appeal vector is then input into the intent recognition model to obtain the probability distribution of intent categories. The intent of the target user is determined based on the probability distribution of intent categories.

Images