Dialogue intention analysis method and device, equipment, storage medium and program product

Abstract

The invention discloses a dialogue intention analysis method and device, equipment, a storage medium and a program product, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining at least one dialogue statement sequence, and carrying out the intention classification analysis of dialogue statements in the at least one dialogue statement sequence, obtaining each type of intention and at least one dialogue statement classified into each type of intention; based on the time sequence of each dialogue statement in the corresponding dialogue statement sequence and the class intention corresponding to each dialogue statement, determining a jump relationship between different class intentions; according to the jump relation between the different types of intentions, an intention relation graph is constructed, the intention relation graph comprises a plurality of nodes and a plurality of directed edges, the nodes represent the types of intentions, and the directed edges represent the jump relation between the different types of intentions; and generating an intention analysis result based on the intention relation graph. The dialogue intention analysis efficiency can be improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a conversation intention analysis method, device, equipment, storage medium and program product. Background Art

[0002] In customer service scenarios, whether interacting with customers through human customer service or intelligent conversational robots, a huge conversation data set will be accumulated. As the complexity and diversity of conversation data continue to increase, conversation intent analysis has become a research field that has attracted much attention.

[0003] Since conversation data is usually diverse, the analysis performance of current conversation intent analysis techniques is often affected, resulting in low efficiency in intent analysis of conversation data resources.

[0004] The above contents are only used to assist in understanding the technical solution of the present application and do not constitute an admission that the above contents are prior art. Summary of the invention

[0005] The main purpose of this application is to provide a conversation intention analysis method, device, equipment, storage medium and program product to improve the efficiency of intention analysis of conversation data resources.

[0006] To achieve the above objectives, the present application proposes a method for analyzing conversation intentions, which includes:

[0007] Acquire at least one dialogue sentence sequence, and perform intent classification analysis on the dialogue sentences in the at least one dialogue sentence sequence to obtain each type of intent and at least one dialogue sentence classified into each type of intent;

[0008] Determine the jump relationship between different class intentions based on the time sequence of each of the dialogue sentences in the corresponding dialogue sentence sequence and the class intentions corresponding to each of the dialogue sentences;

[0009] According to the jump relationship between the different types of intent, construct an intent relationship graph, wherein the intent relationship graph includes a plurality of nodes and a plurality of directed edges, wherein the nodes represent the class intents, and the directed edges represent the jump relationship between the different types of intents;

[0010] Based on the intention relationship graph, an intention analysis result is generated.

[0011] In one embodiment, the step of performing intent classification analysis on the dialogue sentences in the at least one dialogue sentence sequence to obtain each type of intent and at least one dialogue sentence classified into each type of intent includes:

[0012] Performing semantic clustering on all dialogue sentences included in the at least one dialogue sentence sequence to obtain sentence clusters, each of the sentence clusters including at least one dialogue sentence;

[0013] Extracting a dialogue sentence from each of the sentence clusters as a representative sentence;

[0014] Each of the representative sentences is input into the intention recognition model to obtain the intention corresponding to each of the representative sentences, and the intention corresponding to the corresponding representative sentence is set as the class intention corresponding to the corresponding sentence cluster, and the intention of each of the dialogue sentences in the corresponding sentence cluster is uniformly classified into the corresponding class intention.

[0015] In one embodiment, the step of determining the jump relationship between different class intentions based on the time sequence of each of the dialogue sentences in the corresponding dialogue sentence sequence and the class intentions corresponding to each of the dialogue sentences comprises:

[0016] For any of the dialogue sentence sequences, converting each of the dialogue sentences in the dialogue sentence sequence into a class intention corresponding to each of the dialogue sentences, to obtain a class intention sequence, wherein the class intention sequence includes a plurality of class intentions, and the plurality of class intentions are sorted according to a time sequence of the corresponding dialogue sentences in the dialogue sentence sequence;

[0017] Determine that there is a jump relationship between two adjacent class intents in the class intent sequence.

[0018] In one embodiment, the directed edge has characteristic information of jump frequency between class intents, and the step of generating an intent analysis result based on the intent relationship graph includes:

[0019] Traversing each of the directed edges in the intention relationship graph and the jump frequencies corresponding to each of the directed edges, and selecting the directed edges whose corresponding jump frequencies are greater than a preset frequency threshold as high-frequency directed edges;

[0020] Identify the starting node and the target node connected by the high-frequency directed edge, and obtain the class intent corresponding to the starting node as the starting class intent, and the intent corresponding to the target node as the target class intent;

[0021] Generate a high-frequency question-answer intent pair based on the starting class intent and the target class intent;

[0022] Based on each of the high-frequency question-answer intention pairs, the intention analysis result is obtained.

[0023] In one embodiment, the node has characteristic information of the frequency of occurrence of the class intent, and the step of generating the intent analysis result based on the intent relationship graph includes:

[0024] Traversing each of the nodes in the intention relationship graph, and selecting corresponding nodes whose occurrence frequency is greater than a preset frequency threshold as high-frequency nodes;

[0025] Identify an upstream node and / or a downstream node connected to the high-frequency node, and obtain a class intent corresponding to the high-frequency node, a class intent corresponding to the upstream node, and / or a class intent corresponding to the downstream node;

[0026] The intent analysis result is generated based on the class intent corresponding to the high-frequency node, the class intent corresponding to the upstream node and / or the class intent corresponding to the downstream node.

[0027] In one embodiment, the step of generating an intention analysis result based on the intention relationship graph includes:

[0028] Traversing each of the nodes in the intention relationship graph, and selecting nodes whose number of directed edges connected to the nodes is greater than a preset number as cross nodes;

[0029] Obtaining the class intentions corresponding to each of the intersection nodes;

[0030] The intent analysis result is generated based on the class intents corresponding to each of the intersection nodes.

[0031] In addition, to achieve the above purpose, the present application also proposes a conversation intention analysis device, including:

[0032] An intention classification module, used to obtain at least one dialogue sentence sequence, and perform intention classification analysis on the dialogue sentences in the at least one dialogue sentence sequence to obtain each type of intention and at least one dialogue sentence classified into each type of intention;

[0033] A jump relationship determination module, used to determine the jump relationship between different class intentions based on the time sequence of each of the dialogue sentences in the corresponding dialogue sentence sequence and the class intentions corresponding to each of the dialogue sentences;

[0034] A construction module, used to construct an intent relationship graph according to the jump relationship between the different types of intents, wherein the intent relationship graph includes a plurality of nodes and a plurality of directed edges, wherein the nodes represent class intents, and the directed edges represent the jump relationship between the different types of intents;

[0035] The analysis module generates an intent analysis result based on the intent relationship diagram.

[0036] In addition, to achieve the above-mentioned purpose, the present application also proposes a conversation intention analysis device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, and the computer program is configured to implement the steps of the conversation intention analysis method described above.

[0037] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by the processor, the steps of the conversation intention analysis method described above are implemented.

[0038] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the conversation intention analysis method described above.

[0039] One or more technical solutions proposed in this application have at least the following technical effects:

[0040] By acquiring at least one dialogue sentence sequence and performing intent classification analysis on the dialogue sentences in the at least one dialogue sentence sequence, each type of intent and at least one dialogue sentence classified into each type of intent are obtained, and the intents of similar dialogue sentences can be classified into the same type of intent, so that the boundaries between different intents are clear, and redundant intents are effectively reduced. Based on the timing of each dialogue sentence in the corresponding dialogue sentence sequence and the class intent corresponding to each dialogue sentence, the jump relationship between different types of intent is determined. Then, according to the jump relationship between different types of intent, an intent relationship graph is constructed. The intent relationship graph includes multiple nodes and multiple directed edges. The nodes represent each type of intent, and the directed edges represent the jump relationship between different types of intent. The jump relationship between class intents can be mined, and a graph structure is used to store class intents and the jump relationship between class intents. Complex query and analysis operations can be supported. Then, based on the intent relationship graph, an intent analysis result is generated. Since the graph structure supports complex query and analysis operations, it is conducive to in-depth mining of potential relationships and patterns between class intents, and the efficiency of dialogue intent analysis can be improved. In summary, the present application can effectively reduce the increase in the complexity of intent relationship modeling caused by the existence of a large number of similar intents by classifying the intents of similar dialogue sentences into the same category of intent and constructing an intent relationship graph with category intents as nodes. At the same time, since the constructed intent relationship graph has an efficient query and analysis mechanism, it greatly promotes the statistical analysis of intent, and ultimately improves the efficiency of intent analysis as a whole. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0042] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0043] Figure 1 A flowchart of an embodiment of a method for analyzing conversation intentions of the present application is provided;

[0044] Figure 2 A schematic diagram of the intended relationship diagram involved in this application;

[0045] Figure 3 Schematic diagram of the device structure of the hardware operating environment involved in the conversation intention analysis method in the embodiment of the present application.

[0046] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0047] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0048] The main solution of the embodiment of the present application is: obtaining at least one dialogue statement sequence, and performing intent classification analysis on the dialogue statements in the at least one dialogue statement sequence to obtain each type of intent and at least one dialogue statement classified into each type of intent; determining the jump relationship between different types of intent based on the timing of each dialogue statement in the corresponding dialogue statement sequence and the class intent corresponding to each dialogue statement; constructing an intent relationship graph according to the jump relationship between different types of intent, the intent relationship graph including multiple nodes and multiple directed edges, the nodes representing the class intent, and the directed edges representing the jump relationship between different types of intent; generating an intent analysis result based on the intent relationship graph.

[0049] Existing customer service scenarios generate a large number of conversation data sets, which are the basic data for conversation intent analysis. However, since conversation data is usually highly complex and diverse, the statistical analysis performance of current conversation intent analysis technologies is often affected, resulting in low efficiency in intent statistical analysis of conversation data resources.

[0050] The present application provides a solution, which can effectively reduce the increase in the complexity of intent relationship modeling caused by the existence of a large number of similar intents by classifying the intents of similar dialogue sentences into the same category of intent and constructing an intent relationship graph with category intents as nodes. At the same time, since the constructed intent relationship graph has an efficient query and analysis mechanism, it greatly promotes the performance of intent statistical analysis, and ultimately improves the efficiency of intent analysis as a whole.

[0051] It should be noted that the execution subject of this embodiment may be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, a conversation intention analysis device, etc. The following takes the conversation intention analysis device as an example to illustrate this embodiment and the following embodiments.

[0052] See also Figure 1 , Figure 1 This is a flow chart of a method for analyzing conversation intentions according to an embodiment of the present application. In this embodiment, the method for analyzing conversation intentions includes steps S10 to S40:

[0053] Step S10, obtaining at least one dialogue sentence sequence, and performing intent classification analysis on the dialogue sentences in the at least one dialogue sentence sequence to obtain each type of intent and at least one dialogue sentence classified into each type of intent.

[0054] Specifically, the conversation intention analysis device can pre-collect the original conversation data generated by a target user and customer service in multiple historical conversations. A conversation refers to a round of question-and-answer interaction or multiple rounds of question-and-answer interaction between customer service and users around a specific task. Customer service can be a manual customer service or an intelligent customer service robot. The original conversation data can come from online customer service conversation data or from voice call data. If it is voice call data, the voice call data is converted into text conversation data, and the collected multiple original conversation data are cleaned and other operations are performed to obtain a conversation data set. A conversation can generate multiple conversation sentences, and multiple conversation sentences can be sorted into a conversation sentence sequence according to the timestamp of the conversation sentence.

[0055] In this embodiment, the intention refers to the intention of the dialogue, which is used to indicate the purpose of the dialogue. In this embodiment, the intention may include an intention action and an intention content. The intention action refers to the type of behavior expressed in the dialogue sentence, and the intention content is the concretization or supplement of the intention action, which is used to further clarify the specific direction or demand of the dialogue discourse. For example, the dialogue sentence is "I want to check the account balance", the corresponding intention action can be query, and the corresponding intention content is the account balance, then the final intention can be "query-account balance". Intent classification analysis refers to classifying sentences with similar intentions into the same category by identifying the similarity of intentions in dialogue sentences, and giving this category a unified class intention name. Class intention represents the core purpose or action direction shared by a group of dialogue sentences with similar intentions. For example, the two intentions of "booking-movie tickets" and "purchasing-movie tickets" can be classified into the intention category of "purchasing / booking-movie tickets" due to their similarity in core purpose, where "purchasing-movie tickets" is the class intention corresponding to these two intentions.

[0056] In this embodiment, the conversation intention analysis device can use a text clustering method to perform text clustering on all conversation sentences, and cluster semantically similar conversation sentences into one category to obtain each sentence cluster, and select a conversation sentence from each sentence cluster as a representative sentence corresponding to the corresponding sentence cluster, and identify the intention of the representative sentence in the corresponding sentence cluster. It can be understood that the intentions of semantically similar conversation sentences are often similar or the same, so the class intention of all conversation sentences in each sentence cluster can be set to the intention of the representative sentence in the corresponding sentence cluster, so as to classify the intention of each conversation sentence into the corresponding class intention.

[0057] Step S20, based on the timing of each dialogue sentence in the corresponding dialogue sentence sequence and the class intention corresponding to each dialogue sentence, determine the jump relationship between different class intentions.

[0058] Specifically, the dialogue intention analysis device can use the timing of each dialogue sentence in the corresponding dialogue sentence sequence as the timing of each type of intention in the corresponding dialogue sentence sequence, and based on the timing of each type of intention in the corresponding dialogue sentence sequence, obtain the jump relationship between the various types of intention extracted from the corresponding dialogue sentence sequence, and aggregate the jump relationship between the various types of intention extracted from at least one dialogue sentence sequence to obtain the jump relationship between all types of intentions contained in at least one dialogue statement sequence.

[0059] Step S30, construct an intent relationship graph according to the jump relationship between different types of intents, the intent relationship graph includes multiple nodes and multiple directed edges, the nodes represent class intents, and the directed edges represent the jump relationship between different types of intents.

[0060] After obtaining each class intent, the conversation intention analysis device creates a node for each class intent, each node represents a specific class intent, and stores the feature information of the class intent. The feature information of the class intent may include the class intent name, class intent description information, at least one conversation sentence associated with the class intent, and the frequency of occurrence of the class intent. The frequency of occurrence of the class intent can be obtained by calculating the ratio of the frequency of occurrence of the class intent in multiple historical conversations to the total frequency of occurrence of all class intents in multiple historical conversations. In addition, the conversation intention analysis device can construct a directed edge between corresponding nodes based on the jump relationship between different class intents to indicate the jump direction between the corresponding class intents. At the same time, the conversation intention analysis device can store the jump relationship between the class intents represented by the directed edge (also called the jump relationship of the class intent pair), and store the feature information of the class intent pair, and the feature information of the class intent pair includes the jump frequency between the two class intents contained in the class intent pair. The jump frequency between the two class intents contained in a class intent pair can be obtained by the ratio of the cumulative occurrence frequency of the class intent pair to the sum of the cumulative occurrence frequencies of all class intent pairs, thereby quantifying the conversion activity between different class intents.

[0061] As an example, see Figure 2 , Figure 2 The intent relationship diagram involved in the embodiment of the present application is shown. Assuming that the jump relationship between class intents includes the jump relationship from class intent A to class intent B, the jump relationship from class intent A to class intent C, the jump relationship from class intent B to class intent D, the jump relationship from class intent D to class intent F, and the jump relationship from class intent C to class intent E, and assuming that the intent nodes corresponding to class intent A, class intent B, class intent C, class intent D, class intent E, and class intent F are node 1, node 2, node 3, node 4, node 5, and node 6, respectively, then the following can be obtained: Figure 2 Intent diagram shown.

[0062] Step S40, generating intent analysis results based on the intent relationship graph.

[0063] In a feasible implementation, after obtaining the intention relationship graph, the conversation intention analysis device can output the intention analysis results including the intentions corresponding to each node in the intention relationship graph and the feature information of the corresponding intentions, as well as the jump relationships between different types of intentions.

[0064] In another feasible implementation, after obtaining the intention relationship graph, the conversation intention analysis device can use a graph traversal algorithm to traverse each node and each directed edge in the intention relationship graph. During the traversal process, the class intent corresponding to each node and the feature information of the class intent are traversed, as well as the jump relationship of the corresponding class intent pair represented by each directed edge and the feature information of the corresponding class intent pair are traversed. Then, the nodes in the intention relationship graph or the paths in the intention relationship graph are analyzed from multiple dimensions. For example, it may include but is not limited to high-frequency intentions in the intention relationship graph, frequently occurring jump paths, and the intent information represented by the nodes with the most connected directed edges, and the intent analysis results are obtained based on the statistical results obtained from the multi-dimensional analysis.

[0065] This embodiment uses the above scheme to obtain at least one dialogue sentence sequence, and perform intent classification analysis on the dialogue sentences in the at least one dialogue sentence sequence to obtain each type of intent and at least one dialogue sentence classified into each type of intent. The intents of similar dialogue sentences can be classified into the same type of intent, so that the boundaries between different intents are clear, and redundant intents are effectively reduced. Based on the timing of each dialogue sentence in the corresponding dialogue sentence sequence and the class intents corresponding to each dialogue sentence, the jump relationship between different types of intents is determined. Then, an intent relationship graph is constructed. The intent relationship graph includes multiple nodes and multiple directed edges. The nodes represent each type of intent, and the directed edges represent the jump relationship between different types of intents. The jump relationship between class intents can be mined, and a graph structure is used to store class intents and the jump relationship between class intents. Complex query and analysis operations can be supported. Then, based on the intent relationship graph, an intent analysis result is generated. Since the graph structure supports complex query and analysis operations, it is conducive to in-depth mining of potential relationships and patterns between class intents, and the efficiency of dialogue intent analysis can be improved. In summary, the present application can effectively reduce the increase in the complexity of intent relationship modeling caused by the existence of a large number of similar intents by classifying the intents of similar dialogue sentences into the same category of intent and constructing an intent relationship graph with category intents as nodes. At the same time, since the constructed intent relationship graph has an efficient query and analysis mechanism, it greatly promotes the statistical analysis of intent, and ultimately improves the efficiency of intent analysis as a whole.

[0066] In a feasible implementation manner, in the above step S10, performing intent classification analysis on the dialogue sentences in at least one dialogue sentence sequence to obtain each type of intent and at least one dialogue sentence classified into each type of intent may include steps S101 to S103:

[0067] Step S101 , semantically clustering all dialogue sentences contained in at least one dialogue sentence sequence to obtain sentence clusters, each sentence cluster including at least one dialogue sentence.

[0068] Among them, semantic clustering refers to grouping dialogue sentences into different clusters according to their semantic content, and dialogue sentences in the same cluster are semantically similar.

[0069] Specifically, the conversation intention analysis device can use word embedding to convert the conversation sentences into semantic vectors, and cluster the semantic vectors through a clustering algorithm (such as DBSCAN or hierarchical clustering) to obtain clustering results. The clustering results may include the sentence clusters into which each conversation sentence is divided.

[0070] Step S102: extract a dialogue sentence from each sentence cluster as a representative sentence.

[0071] Specifically, the conversation intention analysis device can use the mean-square error (MSE) distance formula or other distance formulas to calculate the distance between each semantic vector in each sentence cluster and the corresponding cluster center, and select the conversation sentence with the smallest distance between the corresponding semantic vector and the cluster center as the representative sentence. Alternatively, the conversation intention analysis device can randomly select a conversation sentence in the corresponding sentence cluster as the representative sentence corresponding to the corresponding sentence cluster. By selecting representative sentences, redundant data in subsequent intention recognition tasks is reduced.

[0072] Step S103, input each representative sentence into the intention recognition model respectively, obtain the intention corresponding to each representative sentence, set the intention corresponding to the corresponding representative sentence as the class intention corresponding to the corresponding sentence cluster, and uniformly classify the intentions of all dialogue sentences in the corresponding sentence cluster into the corresponding class intention.

[0073] The intent recognition model may be a pre-trained model for identifying the intent labels of dialogue sentences. After obtaining each representative sentence, the dialogue intent analysis device inputs each representative sentence into the intent recognition model to obtain the intent corresponding to each representative sentence, and then sets the intent corresponding to the corresponding representative sentence as the class intent corresponding to the corresponding sentence cluster, and uniformly classifies the intents of all dialogue sentences in the corresponding sentence cluster into the corresponding class intent.

[0074] In the above implementation, a large number of dialogue sentences are classified into various sentence clusters through semantic clustering, and representative sentences are selected to identify intent, which reduces the number of dialogue sentences that need to be processed separately and improves data processing efficiency. It is particularly suitable for scenarios that need to process a large amount of dialogue data with high repetitiveness. In addition, similar dialogue sentences are clustered into a class intent, so that the boundaries between intent categories are clearer, overlapping and redundant intent categories are reduced, and the complexity of subsequent intent relationship construction can be reduced.

[0075] In a feasible implementation manner, the above step S20 may include steps S201 to S202:

[0076] Step S201, for any dialogue sentence sequence, convert each dialogue sentence in the dialogue sentence sequence into a class intention corresponding to each dialogue sentence, and obtain a class intention sequence, where the class intention sequence includes multiple class intentions, and the multiple class intentions are sorted according to the time sequence of the corresponding dialogue sentences in the dialogue sentence sequence.

[0077] In this embodiment, the conversation intention analysis device converts each conversation sentence in the conversation sequence into a corresponding class intention, and a corresponding class intention sequence can be obtained. The multiple class intentions in the class intention sequence are sorted according to the time sequence of the corresponding conversation sentences in the original conversation sequence, that is, each intention in the intention sequence corresponds to its position in the original conversation, so as to reflect the changes and flow of the conversation intention in the corresponding conversation. A jump relationship is formed between any two adjacent class intentions in the class intention sequence, and this relationship can be used to characterize the jump direction of the conversation logic, that is, the class intention in the front will jump to the class intention in the back.

[0078] Step S202, determining whether there is a jump relationship between two adjacent class intents in the class intent sequence.

[0079] Specifically, the conversation intention analysis device can initialize an empty class intention jump relationship set, and then traverse the jump relationships between multiple pairs of class intentions extracted from each class intention sequence, and add the jump relationships between each pair of class intentions extracted from each class intention sequence to the class intention relationship set. During the traversal process, if it is found that the jump relationship between any pair of class intentions exists in the class intention jump relationship set, the jump relationship between the pair of class intentions is skipped, and the operation of adding the jump relationship between the pair of class intentions to the class intention relationship set is not performed, so as to achieve deduplication of jump relationships between the same class intentions, which can effectively avoid the repeated addition of jump relationships between class intentions, save storage space, and ensure the uniqueness of the jump relationships between the finally generated class intentions.

[0080] Based on the above embodiments, several implementation schemes for analyzing the intent relationship graph from different dimensions are provided below. Specifically, the intent analysis elements in the intent relationship graph can be mined from the following perspectives, including at least one of the target conversation intent trajectory, high-frequency class intent, high-frequency intent question-answer pairs, cross-class intent, and emotional class intent, and the intent analysis results can be generated based on the mined intent analysis elements.

[0081] The target conversation intention trajectory refers to the trajectory formed by the class intentions corresponding to each conversation sentence in the target conversation. In specific implementation, the conversation intention analysis device can identify multiple node jump relationships associated with the target conversation identifier in the intention relationship graph, and obtain the target conversation trajectory based on the intentions corresponding to the multiple nodes. For example, assuming that the target conversation identifier is k1, the jump relationships between multiple nodes associated with the k1 identifier can be screened out from the intention relationship graph (a→b) k1 , (b→c) k1 and (c→d) k1 , and further obtain the class intents of the corresponding node identifiers a, b, c, and d. Assuming that the class intents of the nodes identified as a, b, c, and d are A, B, C, and D respectively, the target conversation intent trajectory can be obtained as A→B→C→D. By extracting the target conversation intent trajectory, the conversation intent analysis device can clearly track the user's intent change trajectory during the conversation.

[0082] The high-frequency class intent in the target conversation intention trajectory refers to the high-frequency class intent in the target conversation intention trajectory. Specifically, after identifying the class intent of each conversation sentence, the conversation intention analysis device can calculate the frequency of occurrence of each type of intent, and then use the corresponding intent in the target conversation intention trajectory with a frequency greater than a preset threshold as the high-frequency class intent in the target conversation intention trajectory. By identifying the high-frequency class intent, it is possible to understand which conversations are of high frequency concern to customers and conduct in-depth analysis.

[0083] High-frequency class intent refers to the class intent whose corresponding appearance frequency among all class intents is greater than the preset threshold.

[0084] High-frequency question-answer intent pairs refer to class intent pairs whose corresponding occurrence frequency among all class intents in the intent relationship graph is greater than a preset threshold.

[0085] Cross-class intent refers to a class of intent associated with multiple session identifiers. If a class of intent is associated with multiple session identifiers, it means that the class of intent appears in multiple sessions and is a common intent extracted from multiple sessions. The identification of cross-class intent can reveal the common intent between different sessions, which helps to understand the common needs or problems of users. At the same time, enterprises can discover potential service improvements or market trends, and then formulate more accurate market strategies and service optimization plans.

[0086] Emotional intent refers to the intent that represents emotions. For example, this type of intent may contain positive words that can be used to represent happy or satisfied emotions, or negative words that can be used to represent dissatisfied or frustrated emotions. The introduction of emotional intent enables the conversation intent analysis device to perceive the user's emotional state. This emotional analysis capability helps companies understand the user's satisfaction with the service or product, detect and deal with the user's dissatisfaction in a timely manner, and thus maintain good customer relationships.

[0087] The following provides implementation plans for generating intent analysis results based on high-frequency intent question-answer pairs, high-frequency class intents, and cross-class intents:

[0088] In the first embodiment, an implementation scheme for generating intent analysis results based on high-frequency intent question-answer pairs is provided. Specifically, the above step S40 may include steps S401 to S404:

[0089] Step S401, traverse each directed edge in the intention relationship graph and the jump frequency corresponding to each directed edge, and select the directed edges whose corresponding jump frequencies are greater than a preset frequency threshold as high-frequency directed edges.

[0090] Specifically, the conversation intention analysis device can respond to the first intention analysis instruction and start an intention analysis event for the high-frequency intention question and answer pairs in the intention relationship graph. Among them, the intention analysis instruction can be triggered and generated after the user performs a specific operation on a specific control in the visual operation interface. For example, a list of optional controls for the intention analysis elements is displayed in the visual operation interface. When the user clicks on the optional control corresponding to the high-frequency question and answer intention pair, the intention analysis request is triggered. Alternatively, the intention analysis request can also be a pre-set intention analysis event triggered by the conversation intention analysis device after it recognizes that the intention relationship graph is constructed. The intention analysis event includes at least one sub-event, and each sub-event is pre-defined with a corresponding intention analysis instruction, such as the first intention analysis instruction.

[0091] Specifically, directed edges have characteristic information of jump frequencies between intents. After receiving an intent analysis event for a high-frequency intent question-answer pair in the intent relationship graph, the conversation intent analysis device traverses each directed edge in the intent relationship graph and the jump frequency corresponding to each directed edge, and selects directed edges whose corresponding jump frequencies are greater than a preset frequency threshold as high-frequency directed edges.

[0092] Step S402, identify the starting node and the target node connected by the high-frequency directed edge, and obtain the class intent corresponding to the starting node as the starting class intent, and the class intent corresponding to the target node as the target class intent.

[0093] Specifically, each directed edge in the intention relationship graph corresponds to a unique edge identifier, and each directed edge also stores the identifier of the corresponding associated starting node and the identifier of the target node, so that the starting node and the target node connected to the high-frequency directed edge can be obtained based on the identifier of the starting node and the identifier of the target node. Then, the conversation intention analysis device can obtain the class intent corresponding to the starting node (also called the starting class intent) and the class intent corresponding to the target node (also called the target class intent), so as to identify the intent pair with a high-frequency jump relationship.

[0094] Step S403: Generate high-frequency question-answer intent pairs based on the starting class intent and the target class intent.

[0095] It can be understood that dialogue sentences with jump relationships in time sequence usually constitute question-answer pairs, and the extracted starting class intent and target class intent can constitute high-frequency question-answer intention pairs, and the form of high-frequency question-answer intention pairs can be (starting class intent, target class intent).

[0096] Furthermore, after identifying the high-frequency question-answer intention pairs, the conversation intention analysis device may also highlight the high-frequency question-answer intention pairs in the visualized intention relationship graph, and the highlighting method may be: displaying the nodes corresponding to the high-frequency question-answer intention pairs at a first display brightness, and displaying the nodes corresponding to other types of intentions at a second display brightness. Among them, other types of intentions are other types of intentions in the intention relationship graph except for the high-frequency question-answer intention pairs, and the brightness value of the first display brightness is greater than the brightness value of the second display brightness.

[0097] Step S404, obtaining intent analysis results based on each high-frequency question-answer intent pair.

[0098] Specifically, the conversation intention analysis device can obtain at least one conversation sentence corresponding to each type of intent in the high-frequency question and answer intention pairs, and then output the intention analysis results including the high-frequency question and answer intention pairs, the jump frequency of the high-frequency question and answer intention pairs, and at least one conversation sentence corresponding to each type of intent in the high-frequency question and answer intention pairs, and push it to the front end. At the same time, it can also be displayed in the visualization interface according to the pre-defined display method of each type of intention statistical sub-elements. The intention statistical sub-elements may include multiple categories such as the jump frequency between class intents, the occurrence frequency of class intents, and at least one conversation sentence corresponding to class intents.

[0099] In this implementation, by screening out high-frequency question-answer intent pairs in the intent relationship graph and generating intent analysis results containing feature information of high-frequency question-answer intent pairs, request analysts can identify which questions users are more concerned about, which helps companies better understand customer needs and pain points.

[0100] In the second embodiment, an implementation scheme for generating intent analysis results based on high-frequency intent is provided. Specifically, the above step S40 may include steps S401 ˋ to S403 ˋ:

[0101] Step S401, traverse each of the nodes in the intention relationship graph, and select the corresponding nodes whose occurrence frequency is greater than a preset frequency threshold as high-frequency nodes.

[0102] The preset frequency threshold can be set according to actual conditions.

[0103] Step S402: Identify the upstream node and / or downstream node connected to the high-frequency node, and obtain the class intent corresponding to the high-frequency node, the class intent corresponding to the upstream node, and / or the class intent corresponding to the downstream node.

[0104] Among them, the upstream node refers to the node that serves as the starting node and leads to the high-frequency node in a directed edge, and the downstream node refers to the node to which the high-frequency node jumps in a directed edge. When the high-frequency node is the starting node of any directed edge and is not the target node of any directed edge, the conversation intention analysis device can only identify the downstream node corresponding to the high-frequency node, which may include one or more nodes, and then obtain the intention corresponding to the high-frequency node and the intention corresponding to the downstream node. When the high-frequency node is the target node of any directed edge and is not the starting node of any directed edge, the conversation intention analysis device can only identify the upstream node corresponding to the high-frequency node, which may include one or more nodes, and then obtain the intention corresponding to the high-frequency node and the intention corresponding to the upstream node. When the high-frequency node is the target node of any directed edge and is the starting node of another directed edge, the conversation intention analysis device can identify the upstream node corresponding to the high-frequency node and the downstream node corresponding to the high-frequency node, and then obtain the intention corresponding to the high-frequency node, the intention corresponding to the upstream node, and the intention corresponding to the downstream node.

[0105] The conversation intention analysis device can be based on the correspondence between node identifiers and intents pre-stored in the intention relationship graph, and then obtain the class intent corresponding to the high-frequency node based on the high-frequency node identifier (also called high-frequency class intent), obtain the class intent corresponding to the upstream node based on the upstream node identifier (also called upstream class intent), and obtain the class intent corresponding to the downstream node based on the downstream node identifier (also called upstream class intent).

[0106] Step S403: Generate the intent analysis result based on the class intent corresponding to the high-frequency node, the class intent corresponding to the upstream node and / or the class intent corresponding to the downstream node.

[0107] Specifically, the conversation intention analysis device can generate an intention analysis result including high-frequency class intentions and feature information of high-frequency class intentions, upstream class intentions and feature information of upstream class intentions, and / or downstream class intentions and feature information of downstream class intentions. The intention feature information may include the frequency of occurrence of class intentions and at least one corresponding conversation sentence. The conversation intention analysis device can push the obtained intention analysis results to the front end, and can also display them in the visual interface according to the predefined display methods of various intention statistical sub-elements.

[0108] In this implementation, by screening out high-frequency intents in the intent relationship graph and generating intent analysis results containing feature information of high-frequency intents, request analysts can identify which conversations users are more concerned about, which helps companies better understand customer needs and pain points.

[0109] In the third embodiment, an implementation scheme for generating an intention analysis result based on a cross-category intention is provided. Specifically, the above step S40 may include steps S401 ˋ ˋ to S403 ˋ ˋ:

[0110] Step S401 ˋˋ, traverse each node in the intention relationship graph, and select nodes whose corresponding connected directed edges have a number greater than a preset number as cross nodes.

[0111] The preset number may be 2. When the number of directed edges connecting a node is greater than 2, it indicates that the node exists in multiple trajectories and is an intersection node of multiple trajectories.

[0112] Step S402 ˋˋ, obtaining the class intent corresponding to each intersection node.

[0113] Step S403: Generate intent analysis results based on the class intents corresponding to each intersection node.

[0114] Specifically, the conversation intention analysis device can generate an intention analysis result including each cross-class intention (i.e., the class intention corresponding to the cross node) and feature information of the corresponding class intention. The intention feature information may include the frequency of occurrence of the intention and at least one corresponding conversation sentence. The conversation intention analysis device can push the obtained intention analysis result to the front end, and can also display it in the visual interface according to the predefined display method of each type of intention statistical sub-element.

[0115] Through this implementation, it is possible to identify which dialogues have the highest jump activity, thereby optimizing the dialogue system design for the dialogue.

[0116] The embodiment of the present application further provides a conversation intention analysis device, comprising: an intention classification module, used to obtain at least one conversation sentence sequence, and perform intention classification analysis on the conversation sentences in the at least one conversation sentence sequence, to obtain each type of intention and at least one conversation sentence classified into each type of intention;

[0117] A jump relationship determination module, used to determine the jump relationship between different class intentions based on the time sequence of each of the dialogue sentences in the corresponding dialogue sentence sequence and the class intentions corresponding to each of the dialogue sentences;

[0118] A construction module, used to construct an intent relationship graph according to the jump relationship between the different types of intents, wherein the intent relationship graph includes a plurality of nodes and a plurality of directed edges, wherein the nodes represent each type of intent, and the directed edges represent the jump relationship between the different types of intents;

[0119] The analysis module generates an intent analysis result based on the intent relationship diagram.

[0120] In a feasible implementation manner, the above-mentioned intention classification module is also used for:

[0121] Performing semantic clustering on all dialogue sentences included in at least one dialogue sentence sequence to obtain sentence clusters, each of which includes at least one dialogue sentence;

[0122] Extracting a dialogue sentence from each of the sentence clusters as a representative sentence;

[0123] Each of the representative sentences is input into the intention recognition model to obtain the intention corresponding to each of the representative sentences, and the intention corresponding to the corresponding representative sentence is set as the class intention corresponding to the corresponding sentence cluster, and the intentions of all the dialogue sentences in the corresponding sentence cluster are uniformly classified into the corresponding class intention.

[0124] In a feasible implementation manner, the jump relationship determination module is further used to:

[0125] For any of the dialogue sentence sequences, converting each of the dialogue sentences in the dialogue sentence sequence into a class intention corresponding to each of the dialogue sentences, to obtain a class intention sequence, wherein the class intention sequence includes a plurality of class intentions, and the plurality of class intentions are sorted according to the time sequence of the corresponding dialogue sentences in the dialogue sentence sequence, and there is a jump relationship between adjacent class intentions;

[0126] The jump relations between the class intentions extracted from the plurality of dialogue sentence sequences are merged according to a preset merging rule to obtain the jump relations between all the different class intentions extracted from the at least one dialogue sentence sequence.

[0127] In a feasible implementation manner, the above analysis module is also used for:

[0128] Traversing each of the directed edges in the intention relationship graph and the jump frequencies corresponding to each of the directed edges, and selecting directed edges whose corresponding jump frequencies are greater than a preset frequency threshold as high-frequency directed edges;

[0129] Identify the starting node and the target node connected by the high-frequency directed edge, and obtain the class intent corresponding to the starting node as the starting class intent, and the intent corresponding to the target node as the target class intent;

[0130] Generate a high-frequency question-answer intent pair based on the starting class intent and the target class intent;

[0131] Based on each of the high-frequency question-answer intention pairs, the intention analysis result is obtained.

[0132] In a feasible implementation manner, the above analysis module is also used for:

[0133] Traversing each of the nodes in the intention relationship graph, and selecting corresponding nodes whose occurrence frequency is greater than a preset frequency threshold as high-frequency nodes;

[0134] Identify an upstream node and / or a downstream node connected to the high-frequency node, and obtain a class intent corresponding to the high-frequency node, a class intent corresponding to the upstream node, and / or a class intent corresponding to the downstream node;

[0135] The intent analysis result is generated based on the class intent corresponding to the high-frequency node, the class intent corresponding to the upstream node and / or the class intent corresponding to the downstream node.

[0136] In a feasible implementation manner, the above analysis module is also used for:

[0137] Traversing each of the nodes in the intention relationship graph, and selecting nodes whose number of directed edges connected to the nodes is greater than a preset number as cross nodes;

[0138] Obtaining the class intentions corresponding to each of the intersection nodes;

[0139] The intent analysis result is generated based on the class intents corresponding to each of the intersection nodes.

[0140] Compared with the prior art, the beneficial effects of the conversation intention analysis device provided in the embodiment of the present application are the same as the beneficial effects of the conversation intention analysis method provided in the above embodiment, and other technical features in the conversation intention analysis device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0141] An embodiment of the present application provides a language conversation intention analysis device, which may include: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the conversation intention analysis method in the above-mentioned embodiment one.

[0142] Reference below Figure 3 , which shows a schematic diagram of the structure of a language conversation intention analysis device suitable for implementing the embodiment of the present application. The language conversation intention analysis device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 3 The language conversation intention analysis device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0143] like Figure 3As shown, the language conversation intention analysis device may include a processing device 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the language conversation intention analysis device are also stored. The processing device 1001, ROM1002, and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the language conversation intention analysis device to communicate with other devices wirelessly or by wire to exchange data. Although the figure shows a language conversation intention analysis device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or have alternatively.

[0144] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0145] Compared with the prior art, the beneficial effects of the conversation intention analysis device provided in the embodiment of the present application are the same as the beneficial effects of the conversation intention analysis method provided in the above embodiment, and other technical features in the conversation intention analysis device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0146] It should be understood that the various parts disclosed in the embodiments of the present application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0147] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0148] An embodiment of the present application provides a computer-readable storage medium having computer-readable program instructions (ie, a computer program) stored thereon, wherein the computer-readable program instructions are used to execute the conversation intention analysis method in the above-mentioned embodiment.

[0149] The computer-readable storage medium provided in the embodiment of the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM: Random Access Memory), a read-only memory (ROM: Read Only Memory), an erasable programmable read-only memory (EPROM: Erasable Programmable Read Only Memory or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM: CD-Read Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency: Radio Frequency), etc., or any suitable combination of the above.

[0150] The computer-readable storage medium may be included in the conversation intention analysis device; or may exist independently without being assembled into the conversation intention analysis device.

[0151] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the language dialogue intention analysis device, the dialogue intention analysis device executes: obtaining at least one dialogue statement sequence, and performing intent classification analysis on the dialogue statements in the at least one dialogue statement sequence to obtain each type of intent and at least one dialogue statement classified into each type of intent; determining the jump relationship between different types of intent based on the timing of each dialogue statement in the corresponding dialogue statement sequence and the type of intent corresponding to each dialogue statement; constructing an intent relationship graph, the intent relationship graph including multiple nodes and multiple directed edges, the nodes representing each type of intent, and the directed edges representing the jump relationship between different types of intent; generating an intent analysis result based on the intent relationship graph.

[0152] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0153] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0154] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0155] The readable storage medium provided in the embodiment of the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned conversation intent analysis method. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in the embodiment of the present application are the same as the beneficial effects of the conversation intent analysis method provided in the above-mentioned embodiment, which will not be repeated here.

[0156] An embodiment of the present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned conversation intention analysis method when executed by a processor.

[0157] Compared with the prior art, the beneficial effects of the computer program product provided in the embodiment of the present application are the same as the beneficial effects of the conversation intention analysis method provided in the above-mentioned embodiment, which will not be elaborated here.

[0158] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A method for analyzing conversation intention, characterized in that: The method comprises: Acquire at least one dialogue sentence sequence, and perform intent classification analysis on the dialogue sentences in the at least one dialogue sentence sequence to obtain each type of intent and at least one dialogue sentence classified into each type of intent; Determine the jump relationship between different class intentions based on the time sequence of each of the dialogue sentences in the corresponding dialogue sentence sequence and the class intentions corresponding to each of the dialogue sentences; According to the jump relationship between the different types of intent, construct an intent relationship graph, wherein the intent relationship graph includes a plurality of nodes and a plurality of directed edges, wherein the nodes represent the class intents, and the directed edges represent the jump relationship between the different types of intents; Based on the intention relationship graph, an intention analysis result is generated.

2. The method according to claim 1, characterized in that The step of performing intent classification analysis on the dialogue sentences in the at least one dialogue sentence sequence to obtain each type of intent and at least one dialogue sentence classified into each type of intent comprises: Performing semantic clustering on all dialogue sentences included in the at least one dialogue sentence sequence to obtain sentence clusters, each of the sentence clusters including at least one dialogue sentence; Extracting a dialogue sentence from each of the sentence clusters as a representative sentence; Each of the representative sentences is input into the intention recognition model to obtain the intention corresponding to each of the representative sentences, and the intention corresponding to the corresponding representative sentence is set as the class intention corresponding to the corresponding sentence cluster, and the intention of each of the dialogue sentences in the corresponding sentence cluster is uniformly classified into the corresponding class intention.

3. The method according to claim 1, characterized in that The step of determining the jump relationship between different class intentions based on the time sequence of each of the dialogue sentences in the corresponding dialogue sentence sequence and the class intentions corresponding to each of the dialogue sentences comprises: For any of the dialogue sentence sequences, converting each of the dialogue sentences in the dialogue sentence sequence into a class intention corresponding to each of the dialogue sentences, to obtain a class intention sequence, wherein the class intention sequence includes a plurality of class intentions, and the plurality of class intentions are sorted according to the time sequence of the corresponding dialogue sentences in the dialogue sentence sequence; Determine that there is a jump relationship between two adjacent class intents in the class intent sequence.

4. The method according to claim 1, characterized in that The directed edge has characteristic information of jump frequency between class intentions, and the step of generating an intention analysis result based on the intention relationship graph includes: Traversing each of the directed edges in the intention relationship graph and the jump frequencies corresponding to each of the directed edges, and selecting the directed edges whose corresponding jump frequencies are greater than a preset frequency threshold as high-frequency directed edges; Identify the starting node and the target node connected by the high-frequency directed edge, and obtain the class intent corresponding to the starting node as the starting class intent, and the intent corresponding to the target node as the target class intent; Generate a high-frequency question-answer intent pair based on the starting class intent and the target class intent; Based on each of the high-frequency question-answer intention pairs, the intention analysis result is obtained.

5. The method according to claim 1, characterized in that The node has characteristic information of the frequency of occurrence of the class intent, and the step of generating the intent analysis result based on the intent relationship graph includes: Traversing each of the nodes in the intention relationship graph, and selecting corresponding nodes whose occurrence frequency is greater than a preset frequency threshold as high-frequency nodes; Identify an upstream node and / or a downstream node connected to the high-frequency node, and obtain a class intent corresponding to the high-frequency node, a class intent corresponding to the upstream node, and / or a class intent corresponding to the downstream node; The intent analysis result is generated based on the class intent corresponding to the high-frequency node, the class intent corresponding to the upstream node and / or the class intent corresponding to the downstream node.

6. The method according to claim 1, characterized in that The step of generating an intention analysis result based on the intention relationship graph includes: Traversing each of the nodes in the intention relationship graph, and selecting nodes whose number of directed edges connected to the nodes is greater than a preset number as cross nodes; Obtaining the class intentions corresponding to each of the intersection nodes; The intent analysis result is generated based on the class intents corresponding to each of the intersection nodes.

7. A conversation intention analysis device, characterized in that: include: An intention classification module, used to obtain at least one dialogue sentence sequence, and perform intention classification analysis on the dialogue sentences in the at least one dialogue sentence sequence to obtain each type of intention and at least one dialogue sentence classified into each type of intention; A jump relationship determination module, used to determine the jump relationship between different class intentions based on the time sequence of each of the dialogue sentences in the corresponding dialogue sentence sequence and the class intentions corresponding to each of the dialogue sentences; A construction module, used to construct an intent relationship graph according to the jump relationship between the different types of intents, wherein the intent relationship graph includes a plurality of nodes and a plurality of directed edges, wherein the nodes represent class intents, and the directed edges represent the jump relationship between the different types of intents; The analysis module generates an intent analysis result based on the intent relationship diagram.

8. A conversation intention analysis device, characterized in that: The conversation intention analysis device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the conversation intention analysis method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the conversation intention analysis method according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the conversation intention analysis method as described in any one of claims 1 to 6 are implemented.

Images