Intended to identify methods, apparatus, devices, storage media, and program products

Abstract

This application relates to an intent determination method, apparatus, device, storage medium, and program product. The method includes: generating a first intent path based on the intent of a first character in a target scene; inserting a vacant intent into the first intent path to obtain a second intent path; identifying intents adjacent to the vacant intents in the second intent path to obtain a first intent identification result for the vacant intents; and determining the intent of a second character in the target scene based on the first intent identification result. This method, by inserting vacant intents into the first intent path corresponding to the intent of the first character in the target scene to obtain the second intent path, and by identifying intents adjacent to the vacant intents in the second intent path, can comprehensively and accurately determine the intent of the second character in the target scene.

Description

Technical Field

[0001] This application relates to the field of intent recognition technology, and in particular to an intent determination method, apparatus, device, storage medium, and program product. Background Technology

[0002] With the rapid development of computer and internet technologies, the ways in which humans and computers interact have become increasingly diverse, leading to the emergence of dialogue systems, which have been rapidly adopted and applied in people's daily lives.

[0003] Currently, with the diversification of information types and lifestyles, people have higher demands for the intelligence of dialogue systems. However, in current dialogue systems, intent recognition is a crucial core element. Yet, current intent recognition methods are typically based on text classification processing technology. Clearly, this traditional approach is insufficient to meet the ever-changing requirements of dialogue systems, resulting in a need to improve the accuracy of current dialogue systems in recognizing user intent. This also makes it difficult for current dialogue systems to capture users' true needs. Summary of the Invention

[0004] Therefore, it is necessary to provide an intent determination method, apparatus, computer device, computer-readable storage medium, and computer program product that can comprehensively and accurately determine the intent of a second character in a target scenario, in order to address the aforementioned technical problems.

[0005] Firstly, this application provides a method for determining intent. The method includes:

[0006] Generate a first intent path based on the first character's intent in the target scene;

[0007] Insert an empty intent into the first intent path to obtain the second intent path;

[0008] Identify the intents adjacent to the empty space intents in the second intent path to obtain the first intent identification result of the empty space intents;

[0009] The intent of the second character in the target scene is determined based on the first intent recognition result.

[0010] Secondly, this application also provides an intent-determining device. The device includes:

[0011] The generation module is used to generate a first intent path based on the first character's intent in the target scene;

[0012] The insertion module is used to insert an empty intent into the first intent path to obtain the second intent path.

[0013] The recognition module is used to identify the intentions adjacent to the empty space intentions in the second intention path, and obtain the first intention recognition result of the empty space intention;

[0014] The first determination module is used to determine the intention of the second character in the target scene based on the first intention recognition result.

[0015] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement some or all of the steps described in any method of the first aspect of the embodiments of this application.

[0016] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements some or all of the steps described in any method of the first aspect of the embodiments of this application.

[0017] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements some or all of the steps described in any method of the first aspect of the embodiments of this application.

[0018] The aforementioned intent determination method, apparatus, computer device, storage medium, and computer program product, in order to comprehensively encompass the intent paths of both the first and second characters in the target scene, generate a first intent path based on the first character's intent in the target scene. To identify the second character's intent based on the first intent path corresponding to the first character's intent in the target scene, a gap intent is inserted into the first intent path to obtain a second intent path. Intentions adjacent to the gap intent in the second intent path are then identified to obtain a first intent identification result for the gap intent. Finally, the second character's intent in the target scene can be accurately determined based on the first intent identification result. Using the method provided in this application embodiment, by inserting a gap intent into the first intent path corresponding to the first character's intent in the target scene to obtain a second intent path, and by identifying intentions adjacent to the gap intent in the second intent path, the second character's intent in the target scene can be comprehensively and accurately determined. Attached Figure Description

[0019] Figure 1 This is a diagram illustrating the application environment of the method intended to be determined in one embodiment;

[0020] Figure 2 This is a flowchart illustrating the intended method for determining an embodiment;

[0021] Figure 3This is a flowchart illustrating the method intended to be determined in another embodiment;

[0022] Figure 4 This is a structural block diagram of an embodiment intended to determine the device;

[0023] Figure 5 This is an internal structural diagram of a computer device in one embodiment;

[0024] Figure 6 This is a diagram of the internal structure of a computer device in another embodiment. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0026] It should be noted that in the following description, the terms "first, second, and third" are used only to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first, second, and third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0027] The intent determination method provided in this application embodiment can be applied to, for example, Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104, or it can be located in the cloud or on another network server.

[0028] The terminal 102 can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, IoT device, or portable wearable device. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices, etc. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices, etc.

[0029] Server 104 can be an independent physical server or a service node in a blockchain system. The service nodes in the blockchain system form a peer-to-peer (P2P) network. The P2P protocol is an application layer protocol that runs on top of the Transmission Control Protocol (TCP).

[0030] In addition, server 104 can also be a server cluster consisting of multiple physical servers, which can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0031] Terminal 102 and server 104 can be connected via Bluetooth, USB (Universal Serial Bus) or network, etc., and this application does not impose any restrictions.

[0032] Currently, the training of dialogue systems primarily relies on Natural Language Processing (NLP) technology, which broadly falls into three categories: semantic similarity processing, text classification, and natural language inference. While natural language inference can be categorized under text classification, its judgment differs significantly from simple text classification. NLP requires assessing the correspondence between semantic premises and the various components of the hypothetical text to further determine the overall implication of the text.

[0033] Specifically, natural language reasoning technology is mainly used to determine the inference relationships between texts, that is, to determine textual entailment relationships. Textual entailment relationships describe the inference relationship between two texts, where one text is the premise text and the other is the hypothesis text. If the hypothesis text can be inferred from the premise text, then the premise text entails the hypothesis text. It can be seen that identifying textual entailment relationships mainly involves determining whether there is an entailment relationship between the premise text and the hypothesis text. For example, the premise text is "A dog is playing fetch in the snow." Simultaneously, there are three hypothesis texts: the first hypothesis text is "An animal is playing with a plastic toy outdoors in the cold," the second hypothesis text is "A cat is washing its face with its front paws," and the third hypothesis text is "A pet is playing fetch with its owner." Clearly, the first hypothesis text can be inferred from the premise text, the second hypothesis text conflicts with the premise text, and the third hypothesis text has neither an entailment relationship nor a conflict relationship with the premise text. Therefore, in this case, the premise text entails the first hypothesis text.

[0034] Currently, intent recognition is a crucial step in the training process of dialogue systems. However, current intent recognition is typically based on text classification technology. Therefore, integrating natural language reasoning technology into the intent recognition stage of dialogue system training would help improve the efficiency of determining the intent of the corresponding role in the corresponding scenario.

[0035] Based on this, in one embodiment, such as Figure 2 As shown, an intent determination method is provided, which can be derived from... Figure 1 The method is executed by a server or terminal, or by a server and terminal working together. Figure 1 Taking the terminal execution in the example, the following steps are included:

[0036] Step 202: Generate a first intent path based on the first character's intent in the target scene.

[0037] The roles can include call center agents and customers. Call center agents can be either outgoing or incoming call agents, and similarly, customers can also be outgoing or incoming call agents. Furthermore, call center agents can include telemarketing agents, mediation agents, collection agents, and follow-up agents, while customers can include borrowers, complainants, and inquirers.

[0038] The first and second roles can have a corresponding relationship. For example, when the first role is a debt collector, the second role can be a lender; when the first role is a mediator, the second role can be a complainant; similarly, the first role can be a complainant and the second role can be a mediator.

[0039] The first role is a role with a known intention, and there can be more than one first role; the second role is a role whose intention needs to be determined, that is, the second role is a role whose intention is unknown.

[0040] The target scenario can be a call scenario. For example, the call scenario could be a follow-up call with a customer after sales, or a follow-up call regarding the result of complaint handling.

[0041] An intent path can refer to the linear flow path of a corresponding character's intent in a target scene. Thus, a first intent path can refer to the linear flow path between the intents of a first character.

[0042] In one embodiment, the terminal determines the first role's intent in the target scenario through the call log between the first role and the second role. The call log may correspond to the target scenario. The call log may be a voice call log, a text call log, or other forms of call log. The call log only needs to record the call information between the first role and the second role; the presentation format of the call log is not limited here.

[0043] For example, when the call record is in the form of a voice call record, the voice call record can be converted into a text call record, and then the intention of the first character in the target scenario can be determined from the text call record.

[0044] In one embodiment, the terminal responds to an intent determination request initiated by a user on the terminal for a second role, thereby obtaining the call records between the first role and the second role.

[0045] The call records may have different role identifiers to indicate that the call record corresponds to the role. In one embodiment, when the terminal obtains the call record between the first role and the second role, it includes: determining and obtaining the call record with a first role identifier corresponding to the first role and a second role identifier corresponding to the second role, so as to obtain the call record between the first role and the second role.

[0046] Optionally, the intent of the first character in the target scene can be stored locally on the terminal, so that the terminal can obtain the intent of the first character in the target scene from the local storage, and then generate the first intent path based on the intent of the first character in the target scene.

[0047] Step 204: Insert an empty intent into the first intent path to obtain the second intent path.

[0048] Inserting a gap intention into the first intention path can be achieved by inserting a gap intention between adjacent intentions of the first role in the first intention path, thereby obtaining a second intention path based on the intentions adjacent to the gap intention and the gap intention itself; furthermore, it can be achieved by inserting a gap intention between adjacent intentions in the first intention path and after the last intention in the first intention path. Based on this, there can be one or two intentions adjacent to the gap intention in the first intention path.

[0049] An intention adjacent to an empty space intention can refer to the intention to the left and right of the empty space intention in the first intention path, relative to the direction of the intention path; the number of intentions adjacent to an empty space intention is two or less. For example, an intention adjacent to an empty space intention located at a non-end position in the first intention path is two intentions, while an intention adjacent to an empty space intention located at the end position in the first intention path is only one intention.

[0050] The last intent in the first intent path refers to the intent that appears at the very end of the first intent path along the intent path direction; the intent path direction can be from left to right along the intent path.

[0051] For example, a first intent path generated based on intents A, B, C, D, and E of the first role is ABCDE, with the intent path direction being A→…→E. Then, intents A and B, B and C, C and D, and D and E are all adjacent intents, and intent E is the last intent in this first intent path. Therefore, by inserting empty intents between adjacent intents in the first intent path and after the last intent in the first intent path, the resulting first intent path with inserted empty intents is: A-empty intent-B-empty intent-C-empty intent-D-empty intent-E-empty intent.

[0052] The second intention path includes the intention of the first role and the intention of the empty space.

[0053] Step 206: Identify the intents adjacent to the empty space intents in the second intent path to obtain the first intent identification result of the empty space intents.

[0054] In this process, identifying the intent adjacent to the empty space intent in the second intent path can be achieved by identifying the content of the two intents located to the left and right of the empty space intent in the second intent path, and then obtaining the first intent identification result of the empty space intent based on the content of the two intents located to the left and right of the empty space intent.

[0055] In one embodiment, a first adjacent intent and a second adjacent intent of the empty space intent are selected in the second intent path; intent recognition is performed on the first adjacent intent to obtain the second intent recognition result of the empty space intent; intent recognition is performed on the second adjacent intent to obtain the third intent recognition result of the second adjacent intent; and the first intent recognition result is determined based on the second intent recognition result and the third intent recognition result.

[0056] The first adjacent intent of the empty space intent can be the intent located to the left of the empty space intent relative to the intent path direction; similarly, the second adjacent intent of the empty space intent can be the intent located to the right of the empty space intent relative to the intent path direction.

[0057] Optionally, identifying the first and / or second adjacent intentions of the empty intention in the second intention path can be achieved by determining the intention recognition method based on the sentence type of the intentions adjacent to the empty intention, and then identifying the intentions adjacent to the empty intention based on the determined intention recognition method to obtain the first intention recognition result of the empty intention; alternatively, it can be achieved by identifying the first and / or second adjacent intentions of the empty intention based on the intention recognition model to obtain the first intention recognition result of the empty intention.

[0058] Optionally, the intent recognition of the first adjacent intent can be performed based on the sentence structure type of the first adjacent intent to obtain the second intent recognition result of the empty space intent.

[0059] Optionally, the intention recognition of the second adjacent intention can be performed based on the intention recognition model to obtain the third intention recognition result of the second adjacent intention.

[0060] Optionally, determining the first intent recognition result based on the second intent recognition result and the third intent recognition result may involve the second intent recognition result playing a dominant role in determining the first intent recognition result, while the third intent recognition result plays a supplementary role. For example, the third intent recognition result may modify the second intent recognition result to determine the first intent recognition result, thereby jointly determining the first intent recognition result that accurately bridges the preceding and following intent meanings.

[0061] In this embodiment, when identifying the intent adjacent to the empty intent in the second intent path and obtaining the first intent identification result of the empty intent, the first and second adjacent intents of the empty intent are selected in the second intent path, and intent identification is performed on the first and second adjacent intents respectively to obtain the second intent identification result of the empty intent and the third intent identification result of the second adjacent intent. Then, the first intent identification result is determined based on the second and third intent identification results. Thus, by jointly determining the second intent identification result and the third intent identification result, a first intent identification result that accurately connects the preceding and following intents in terms of intent meaning is obtained, improving the accuracy of intent determination.

[0062] For example, identifying the intent adjacent to the empty space intent in the second intent path can be done by determining the intent recognition method based on the sentence type of the intent adjacent to the empty space intent, and then identifying the intent adjacent to the empty space intent based on the determined intent recognition method to obtain the first intent recognition result of the empty space intent; or it can be done by identifying the intent adjacent to the empty space intent based on the intent recognition model to obtain the first intent recognition result of the empty space intent.

[0063] Step 208: Determine the intention of the second character in the target scene based on the first intention recognition result.

[0064] The first intent recognition result corresponds to the intent of the empty space. Therefore, the first intent recognition result is actually the intent of the second character in the target scene.

[0065] Optionally, the similarity of the first intent recognition result can be determined, and intent merging processing can be performed based on the similarity, so that the second character's intent in the target scene does not have duplicate intent or highly similar intent.

[0066] In one embodiment, after determining the second character's intent in the target scene, the terminal stores the second character's intent locally. This achieves the transformation of the second character in the target scene from a character with unknown intent to a character with known intent.

[0067] The number of characters in the second role can be more than one.

[0068] In one embodiment, after determining the intention of the second character in the target scene based on the first intention recognition result, the method may further include: determining the intention set of the second character in the target scene based on the intention of the second character in the target scene. The intention set of the second character obtained based on the embodiments of this application can be used to evaluate the accuracy of intention sets of the second character obtained through other means, can also be used as a reference for merging and reasoning about the intentions of other characters, and can also be used for text data augmentation to expand the training dataset of subsequent intention recognition models and improve model performance.

[0069] In the aforementioned intent determination method, to comprehensively encompass the intent paths of both the first and second characters in the target scene, a first intent path is generated based on the first character's intent in the target scene. To identify the second character's intent based on the first intent path corresponding to the first character's intent in the target scene, a second intent path is obtained by inserting a gap intent into the first intent path. Furthermore, the intents adjacent to the gap intents in the second intent path are identified to obtain the first intent identification result for the gap intents. Ultimately, the second character's intent in the target scene can be accurately determined based on the first intent identification result. By employing the method provided in this application embodiment, a second intent path is obtained by inserting a gap intent into the first intent path corresponding to the first character's intent in the target scene, and by identifying the intents adjacent to the gap intents in the second intent path, the second character's intent in the target scene can be comprehensively and accurately determined.

[0070] In one embodiment, if the sentence type of the first adjacent intent is a first preset sentence type, then the target text is selected from the first adjacent intents, and the second intent recognition result includes the target text. The process of determining the sentence type of the first adjacent intent can be implemented through a trained sentence classification model. The trained sentence classification model can be obtained by classifying and training labeled text and sentence types; therefore, the terminal can determine the sentence type of the first adjacent intent based on the trained sentence classification model.

[0071] For example, the first preset sentence type may include yes / no question, yes / no question, and specific question.

[0072] Yes / no question sentences refer to general interrogative sentences. For example, if the first adjacent intent is "Are you Xiaoming himself?", then the sentence type is determined to be a yes / no question sentence based on the first adjacent intent. In this case, one of "yes", "no", "is Xiaoming himself", and "is not Xiaoming himself" can be selected as the target text, and the selected target text is used as the result of the second intent recognition.

[0073] A rhetorical question is a sentence structure that simultaneously presents both the positive (affirmative) and negative (negative) aspects of a situation within a single question. For example, if the first adjacent intent is "Can you repay on time?", then based on the first adjacent intent, the sentence type is determined to be a rhetorical question. In this case, either "Can repay on time" or "Cannot repay on time" can be selected as the target text, and the selected target text is used as the result of the second intent recognition.

[0074] A specific question is a question type that aims to elicit an answer from the person being asked about a point of contention. For example, if the first adjacent intent is "Do you think the moon is round or square?", then based on the first adjacent intent, the sentence type is determined to be a specific question. In this case, either "The moon is round" or "The moon is square" can be selected as the target text, and the selected target text is used as the result of the second intent recognition.

[0075] In this embodiment, when the sentence type of the first adjacent intent is the first preset sentence type, the target text is selected from the first adjacent intent. Thus, the second intent recognition result of the empty space intent includes the target text. It can be seen that determining the second intent recognition result including the target text based on the sentence type of the first adjacent intent can significantly improve the determination efficiency of the second intent recognition result.

[0076] In one embodiment, if the sentence type of the first adjacent intent is the second preset sentence type, the first adjacent intent is input into the semantic feature recognition model for semantic feature recognition to obtain the semantic features in the first adjacent intent; the semantic features whose frequency of occurrence meets the preset frequency condition are selected as key semantic features, wherein the second intent recognition result includes key semantic features.

[0077] The second preset sentence type can be a different sentence type from the first preset sentence type. For example, when the sentence type of the first adjacent intention includes yes / no question, yes / no question, and specific question, then the second preset sentence type is a sentence type other than yes / no question, yes / no question, and specific question.

[0078] A semantic feature recognition model is a model used to identify the smallest semantic units that constitute semantics, that is, a model used to identify semantic features.

[0079] The purpose of inputting the first adjacent intent into the semantic feature recognition model for semantic feature recognition is to perform semantic feature recognition on the first adjacent intent. That is to say, semantic feature recognition is performed on the first adjacent intent based on a semantic feature recognition model capable of performing semantic feature recognition. Semantic feature recognition involves grouping a group of words in the same semantic field together and analyzing, comparing, and describing them from the perspective of semantic features.

[0080] The frequency of a semantic element in the first adjacent intent can be more than once. If the frequency of a certain semantic element is higher, it means that the semantic element has a more important meaning in the first adjacent intent. Therefore, the preset frequency condition can be that the frequency of occurrence is greater than or equal to the first preset frequency. Thus, the terminal selects the semantic element with a frequency of occurrence greater than or equal to the first preset frequency as the key semantic element.

[0081] In this embodiment, when the sentence type of the first adjacent intention is the second preset sentence type, the first adjacent intention is input into the semantic feature recognition model for semantic feature recognition to obtain the semantic features in the first adjacent intention. Finally, the semantic features whose frequency of occurrence meets the preset frequency condition are selected as key semantic features. Thus, the second intention recognition result of the empty intention includes key semantic features. It can be seen that determining the second intention recognition result including key semantic features based on the sentence type of the first adjacent intention can significantly improve the accuracy of the second intention recognition result.

[0082] It can be seen that, based on the first and second preset sentence types, for different sentence types of the first adjacent intention, the second intention identification with accuracy and logic can be finally determined, thereby avoiding the bad situation of inaccurate or illogical reasoning intentions, and thus ensuring that the intentions of the second role in the target scene obtained subsequently are comprehensive and accurate.

[0083] In one embodiment, the second adjacent intent is segmented to obtain the segmented words of the second adjacent intent; through an intent recognition model, the segmented words of the second adjacent intent are feature-extracted to obtain segmented feature vectors and segmented weight vectors; the segmented feature vectors and segmented weight vectors are recognized to obtain the named entities of the second adjacent intent, wherein the third intent recognition result includes the named entities.

[0084] Tokenization is a fundamental step in Natural Language Processing (NLP), involving dividing a text string into smaller tokens. Tokens can be words, phrases, symbols, or any other meaningful basic unit in a language. Tokenization of the second neighboring intent is performed to ultimately obtain named entities for that intent.

[0085] The word segmentation feature vector represents each word in the text as a point in a multi-dimensional space. Each dimension in this multi-dimensional space corresponds to a feature value, and the feature values ​​of each word constitute the word segmentation feature vector.

[0086] The segmentation weight vector is a quantitative representation of the importance or contribution of each segment in the text. The segmentation weight can be determined based on the term frequency of each segment, TF-IDF (Term Frequency-Inverse Document Frequency), or other weighting strategies.

[0087] Named entities are words in text that refer to specific objects or concepts in the real world. Named entities include names of people, places, organizations, times, quantities, or other entity types.

[0088] In this embodiment, the process of obtaining the third intent recognition result of the second adjacent intent involves segmenting the second adjacent intent into words, then extracting features from the segmented words of the second adjacent intent using an intent recognition model to obtain segmentation feature vectors and segmentation weight vectors. Finally, by recognizing the segmentation feature vectors and segmentation weight vectors, the named entities of the second adjacent intent are obtained. Thus, the third intent recognition result includes the named entities. It can be seen that determining the third intent recognition result including the named entities based on the segmentation processing method can significantly improve the accuracy of the third intent recognition result, and thus can more accurately cooperate with the second intent recognition result to determine the first intent recognition result with higher accuracy.

[0089] In one embodiment, the intention of a first character in a target scene is obtained; the intentions of the first character in the target scene are sorted to obtain an intention sequence; and a first intention path is determined based on the intention sequence.

[0090] In one embodiment, the intentions of the first character in the target scene are numbered in m rounds to obtain m sets of numbers, where m is a positive integer; the intentions of the first character in the target scene are sorted according to the intention numbers in the m sets of numbers to obtain m intention sequences; wherein, the intention sequences correspond one-to-one with the second intention paths; based on the first intention recognition results of the empty intentions in the m second intention paths, the m intention groups of the second character in the target scene are determined.

[0091] The intent of the first character in the target scenario can be obtained by the terminal through the call records between the first and second characters.

[0092] Optionally, during the process of numbering the intentions of the first character in the target scene for m rounds, the total frequency of some or all of the intentions of the first character in the target scene in the corresponding number set can be more than once, and the frequency of consecutive occurrences can also be more than once.

[0093] In one embodiment, the intention of the first character in the target scene is processed by numbering in m rounds to obtain m sets of numbers, including: determining the total frequency and consecutive frequency of the first character's intention in the target scene during the numbering process in each round; wherein the total frequency of occurrence is greater than or equal to a second preset frequency, and the consecutive frequency of occurrence is less than or equal to a third preset frequency; based on the total frequency of occurrence and consecutive frequency of occurrence of the intention during the numbering process in each round, m sets of numbers are obtained.

[0094] The total frequency of an intent in a certain round of numbering process refers to the total frequency of the same intent appearing at different positions in the corresponding intent sequence during that round of numbering process. The consecutive frequency of an intent in a certain round of numbering process refers to the frequency of the same intent appearing continuously at a certain position in the corresponding intent sequence during that round of numbering process; a consecutive frequency less than or equal to a third preset frequency means that the frequency of a certain intent of the first role appearing continuously at a certain position cannot exceed the third preset frequency. Intents appearing at different positions in the same intent sequence, as well as intents appearing continuously at a certain position in the same intent sequence, all correspond to intent numbers, and these intent numbers can be different.

[0095] For example, the second preset frequency can be 1 or other values. When the second preset frequency is 1, the first character's intention in the target scene appears at least once in different intention sequences, thereby ensuring that the occurrence of the first character's intention in the target scene can be covered as much as possible. The third preset frequency can be 3 or other values, thereby avoiding the undesirable situation where the frequency of the same intention appearing consecutively is too high, resulting in meaningless intention interactions between the first character and the second character in the target scene.

[0096] Optionally, when the terminal determines the total frequency and consecutive frequency of an intention in a certain round of numbering process, the total frequency and / or consecutive frequency of an intention in different rounds of numbering process can be different.

[0097] For example, if an intent sequence generated based on intent A, intent B, and intent C of the first role is ABBBCB, then the total number of occurrences of intent A, intent B, and intent C are 1, 4, and 1, respectively, and the consecutive number of occurrences of intent A, intent B, and intent C are 1, 3, and 1, respectively.

[0098] In this embodiment, based on the total frequency and consecutive frequency of the intention in the numbering process of each round, the process of obtaining m intention sequences has uncertainty in the total frequency of intention occurrence, uncertainty in the consecutive frequency of intention occurrence, and uncertainty in the position of intention occurrence. As a result, a large number of possible intention sequences can be generated, which can more comprehensively cover the intention interaction between the first role and the second role in the target scene.

[0099] In the process of numbering the intentions of the first character in the target scene in m rounds, the order in which the intentions appear can be different. That is to say, each of the m intention sets can have a different intention numbering method.

[0100] Each intent sequence corresponds one-to-one with a second intent path, meaning that each intent sequence corresponds to a different second intent path. Thus, the number of second intent paths is the same as the number of intent sequences. Specifically, in the embodiments of this application, the number of intent sequences and second intent paths are both m.

[0101] Optionally, the second character's intention set in the target scenario can refer to all the intentions that the second character might have in a single call record with the first character in the target scenario.

[0102] In this embodiment, the intention of the first character in the target scene is obtained, and then numbering is performed for m rounds to obtain m sets of numbers. The first intention of the first character in the target scene is then sorted according to the intention numbers in the m sets of numbers to obtain m intention sequences. Each of the m intention sequences corresponds one-to-one with a second intention path. Therefore, based on the first intention recognition results of the empty intentions in the m second intention paths, m intention groups of the second character in the target scene can be determined. It can be seen that this embodiment, by combining intention numbering and sorting, can obtain m intention sequences that comprehensively cover the intention interactions of the first character in the target scene. Furthermore, it ensures that the subsequently obtained m intention groups of the second character in the target scene also have comprehensive intention interaction information.

[0103] In one embodiment, for each of the m intent groups, a first similarity is determined between the intents in the intent group; in each intent group, the intents whose first similarity reaches a similarity threshold are merged to obtain the intent merging result of each intent group; based on the intent merging result of each intent group, the intent set of the second role in the target scene is determined.

[0104] The first similarity threshold includes intents that are highly similar or identical. The second set of intents of a second character in the target scene includes multiple intents of multiple second characters in the target scene.

[0105] The intent set can include all intents of the corresponding role and the interpretation of each intent. Therefore, the intent set of the second role in the target scenario includes all intents of the second role in the target scenario and the interpretation of each intent. For example, when the second role is a collection agent in a debt collection scenario, the second role's intent set can include the intents of "collecting payment" and "reminding the repayment date," along with their corresponding interpretations. Similarly, when the second role is a follow-up agent in an after-sales customer follow-up scenario, the second role's intent set can include the intent of "following up on product usage experience," along with its corresponding interpretation; and when the second role is a follow-up agent in a complaint handling result follow-up scenario, the second role's intent set can include the intent of "tracking the complaint handling result," along with its corresponding interpretation.

[0106] The purpose of merging intents that reach the first similarity threshold is to reduce the similarity of intents in different intent groups, avoid highly similar or identical intents in the same intent group, and thus obtain an intent merging result without data redundancy.

[0107] In this embodiment, by merging the intentions in each intention group that reach the first similarity threshold, and then determining the intention set of the second character in the target scene based on the intention merging result of each intention group, the accuracy of the final determined intention set of the second character in the target scene can be significantly improved while reducing the data redundancy of the intention set by reducing the intention similarity in each intention group.

[0108] In one embodiment, the number of intent recognition results is n; a second similarity is determined among the n intent recognition results; among the n intent recognition results, the intent recognition results whose second similarity reaches a similarity threshold are merged to obtain a first merged intent recognition result; the intent of the second character in the target scene is determined based on the first merged intent recognition result.

[0109] The second similarity threshold for intent recognition results includes highly similar intent recognition results as well as identical intent recognition results. The first merged intent recognition result refers to an intent recognition result with low or no data redundancy after merging n intent recognition results. The set of intents of the second role in the target scene includes multiple intents of multiple second roles in the target scene.

[0110] The purpose of merging the intent recognition results that reach the second similarity threshold is to reduce the intent similarity in different intent recognition results, avoid highly similar or identical intents in different intent recognition results, and thus avoid the undesirable situation of data redundancy in the intent of the determined second role in the target scene.

[0111] In this embodiment, by merging the intent recognition results that reach the second similarity threshold, and then determining the intent of the second character in the target scene based on the obtained first merged intent recognition result, the accuracy of the final determined intent set of the second character in the target scene can be improved while reducing the data redundancy of the intent set by reducing the intent similarity between different intent recognition results.

[0112] In one embodiment, the similarity of the intentions adjacent to the empty space intention is calculated to obtain a third similarity; among the intentions adjacent to the empty space intention, the intentions that reach the similarity threshold of the third similarity are merged to obtain a second merged intention recognition result; based on the first merged intention recognition result and the second merged intention recognition result, the intention of the second role in the target scene is determined.

[0113] The second merged intent identification result refers to the intent result that has low data redundancy or no data redundancy after merging the intents adjacent to the empty space intent.

[0114] Optionally, the intent features can be obtained by extracting text features from each intent, and then the intent features can be converted into vector form. Then, the similarity between the intents in vector form can be determined by cosine similarity, Euclidean distance or other algorithms used to evaluate vector similarity, so as to determine the first similarity, the second similarity and the third similarity.

[0115] It can be seen that by merging the intentions that reach the first similarity threshold in each intention group, merging the intentions that reach the second similarity threshold, and merging the intentions that reach the third similarity threshold among the intentions adjacent to the empty slot intentions, we can reduce the intention similarity within each intention group and also reduce the intention similarity between different intention recognition results. Furthermore, by continuously reducing the data redundancy that may exist in the intention determination process, we can further avoid the undesirable situation of data redundancy in the final determined intention set of the second role in the target scene.

[0116] In one embodiment, the above-mentioned identification of the intent adjacent to the empty space intent in the second intent path to obtain the first intent identification result of the empty space intent includes: identifying the second merged intent identification result to obtain the first intent identification result of the empty space intent.

[0117] In this embodiment, the intention of the second role in the target scene is determined jointly based on the first merged intention recognition result with low or no data redundancy, and the second merged intention recognition result with low or no data redundancy. It can be seen that the intention of the second role in the target scene ultimately determined in this embodiment also has low or no data redundancy, thus improving data quality by avoiding data redundancy.

[0118] The application process of the above intent determination method is illustrated below with a detailed embodiment. The intent determination method is applied to a terminal, and the specific application process is as follows:

[0119] (1) The process of determining the first intention path

[0120] Obtain the intent of the first character in the target scene;

[0121] The intentions of the first character in the target scene are numbered in m rounds to obtain a set of m numbers, where m is a positive integer;

[0122] Based on the intent numbers in the m sets of numbers, sort the intents of the first character in the target scene to obtain m intent sequences; where each intent sequence corresponds one-to-one with the second intent path;

[0123] The first intent path is determined based on the intent sequence.

[0124] (2) The process of determining the second intention path

[0125] Insert an empty intent into the first intent path to obtain the second intent path;

[0126] (3) The process of determining the first intent recognition result

[0127] In the second intent path, select the first and second adjacent intents of the empty space intent;

[0128] If the sentence type of the first adjacent intent is the first preset sentence type, then the target text is selected from the first adjacent intent, and the second intent recognition result includes the target text; if the sentence type of the first adjacent intent is the second preset sentence type, then the first adjacent intent is input into the semantic feature recognition model for semantic feature recognition to obtain the semantic features in the first adjacent intent; the semantic features whose occurrence frequency meets the preset frequency condition are selected as key semantic features, wherein the second intent recognition result includes key semantic features;

[0129] The second adjacent intent is segmented into words to obtain the word segments of the second adjacent intent; through the intent recognition model, the word segments of the second adjacent intent are feature extracted to obtain the word segmentation feature vector and word segmentation weight vector;

[0130] The word segmentation feature vector and word segmentation weight vector are identified to obtain the named entities of the second adjacent intent, wherein the third intent identification result includes the named entities;

[0131] The first intent recognition result is determined based on the second intent recognition result and the third intent recognition result;

[0132] (4) The process of determining the set of intentions of the second character in the target scene

[0133] Based on the first intent recognition results of the empty intents in the m second intent paths, determine the m intent groups of the second role in the target scene;

[0134] For each of the m intent groups, determine the first similarity between the intents in the intent group;

[0135] Within each intent group, intents that reach the first similarity threshold are merged to obtain the intent merging result for each intent group;

[0136] Based on the intent merging results of each intent group, determine the set of intents of the second character in the target scene.

[0137] In this embodiment, an intent sequence is obtained by numbering and sorting intents in a set to determine the first intent path, thus covering a wide range of possible first intent paths. A second intent path is obtained by inserting empty intents into the first intent path. The first and second adjacent intents of the empty intents are then processed by sentence structure determination and word segmentation to determine the first intent recognition result of the empty intents, ensuring the logicality and accuracy of the first intent recognition result. Based on the first intent world results of the empty intents in the m second intent paths, m intent groups of the second character in the target scene are determined. Intentions with high similarity in each of the m intent groups are merged, and finally, the intent set of the second character in the target scene is determined based on the intent merging results of each intent group. It can be seen that this embodiment ensures that the final determined intent set of the second character in the target scene is comprehensive and accurate.

[0138] In one embodiment, such as Figure 3 As shown, this application provides an intent determination method, which is used to determine the intent of a role in a debt collection scenario. The method comprises... Figure 1 The terminal execution includes the following steps:

[0139] Step 302: In order to determine the set of intentions of the borrower in the collection scenario and improve the dialogue in the collection scenario, the user specifies all the roles involved, namely the collection agent and the borrower, on the terminal applied to the intention determination method provided in the embodiment of this application.

[0140] Step 304: Obtain all calls from the call logs of the collection scenario that only involve the collection agent and the borrower.

[0141] Step 306: Obtain the set of intents of collection agents in the collection scenario.

[0142] Step 308: Generate collection agent intent paths by categorizing the intents in the collection scenario's intent set according to their total frequency of occurrence and location of occurrence.

[0143] Step 310: Hollow out the space between every two intentions in the collection agent's intention path to leave empty intentions for the borrower in the collection agent's intention path.

[0144] Step 312: Based on the intentions of the two collection agents before and after each empty slot intention, generate the first intention recognition result for each empty slot intention to obtain the borrower's intention path.

[0145] Step 314: Merge the intent recognition results in the borrower's intent path to finally determine the borrower's intent in the collection scenario, and then generate an intent set based on the borrower's intent in the collection scenario.

[0146] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0147] Based on the same inventive concept, this application also provides an intent determination apparatus for implementing the intent determination method described above. The solution provided by this apparatus is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more intent determination apparatus embodiments provided below can be found in the limitations of the intent determination method described above, and will not be repeated here.

[0148] In one embodiment, such as Figure 4 As shown, an intent determination device is provided, comprising: a generation module 402, an insertion module 404, an identification module 406, and a first determination module 408, wherein:

[0149] The generation module 402 is used to generate a first intent path based on the intent of the first character in the target scene;

[0150] Insertion module 404 is used to insert an empty intent into the first intent path to obtain the second intent path;

[0151] The recognition module 406 is used to recognize the intentions adjacent to the empty space intentions in the second intention path and obtain the first intention recognition result of the empty space intentions;

[0152] The first determining module 408 is used to determine the intention of the second character in the target scene based on the first intention recognition result.

[0153] In one embodiment, the identification module 406 is further configured to select a first adjacent intent and a second adjacent intent in the second intent path; perform intent identification on the first adjacent intent to obtain a second intent identification result for the empty intent; perform intent identification on the second adjacent intent to obtain a third intent identification result for the second adjacent intent; and determine a first intent identification result based on the second intent identification result and the third intent identification result.

[0154] In one embodiment, the identification module 406 is further configured to select target text from the first adjacent intent when the sentence type of the first adjacent intent is a first preset sentence type, and the second intent identification result includes the target text.

[0155] In one embodiment, the identification module 406 is further configured to input the first adjacent intention into the semantic feature identification model for semantic feature identification if the sentence type of the first adjacent intention is the second preset sentence type, thereby obtaining the semantic features in the first adjacent intention; and select the semantic features whose frequency of occurrence meets the preset frequency condition as key semantic features, wherein the second intention identification result includes key semantic features.

[0156] In one embodiment, the identification module 406 is further configured to perform word segmentation on the second adjacent intent to obtain word segments of the second adjacent intent; extract features from the word segments of the second adjacent intent through the intent recognition model to obtain word segmentation feature vectors and word segmentation weight vectors; identify the word segmentation feature vectors and word segmentation weight vectors to obtain named entities of the second adjacent intent, wherein the third intent recognition result includes named entities.

[0157] In one embodiment, the generation module 402 is further configured to obtain the intention of the first character in the target scene; sort the intention of the first character in the target scene to obtain an intention sequence; and determine the first intention path based on the intention sequence.

[0158] In one embodiment, the generation module 402 is further configured to perform m rounds of numbering on the intentions of the first character in the target scene to obtain m sets of numbers, where m is a positive integer; and sort the intentions of the first character in the target scene according to the intention numbers in the m sets of numbers to obtain m intention sequences; wherein, the intention sequences correspond one-to-one with the second intention paths.

[0159] The aforementioned first determining module 408 is also used to determine the m intention groups of the second role in the target scene based on the first intention recognition results of the empty intentions in the m second intention paths.

[0160] like Figure 4 As shown, in one embodiment, the above-mentioned device further includes a second determining module 410, which is used to determine the first similarity between each intention in each of the m intention groups; merge the intentions that reach the similarity threshold in each intention group to obtain the intention merging result of each intention group; and determine the intention set of the second character in the target scene based on the intention merging result of each intention group.

[0161] In one embodiment, the number of intent recognition results is n; the first determining module 408 is further used to determine the second similarity between the n intent recognition results; among the n intent recognition results, the intent recognition results whose second similarity reaches the similarity threshold are merged to obtain the first merged intent recognition result; and the intent of the second role in the target scene is determined based on the first merged intent recognition result.

[0162] like Figure 4 As shown, in one embodiment, the above-mentioned device further includes a merging module 412, which is used to calculate the similarity of intentions adjacent to empty space intentions to obtain a third similarity; among the intentions adjacent to empty space intentions, the intentions whose third similarity reaches a similarity threshold are merged to obtain a second merged intention recognition result.

[0163] The first determining module 408 is further used to determine the intention of the second character in the target scene based on the first merged intention recognition result and the second merged intention recognition result.

[0164] The modules in the aforementioned intent-determining device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware within or independently of the processor in a computer device, or stored in software within the memory of a computer device, so that the processor can invoke and execute the operations corresponding to each module.

[0165] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 5 As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores intent data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When executed by the processor, the computer program implements an intent determination method.

[0166] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 6As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage medium. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements an intent determination method. The display unit of the computer device is used to form a visually visible image. It can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0167] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0168] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0169] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0170] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0171] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data shall comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0172] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0173] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0174] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. An intent determination method, characterized in that, The method includes: Generate a first intent path based on the first character's intent in the target scene; Insert an empty intent into the first intent path to obtain the second intent path; Identify the intents adjacent to the empty space intent in the second intent path to obtain the first intent identification result of the empty space intent; The intent of the second character in the target scene is determined based on the first intent recognition result.

2. The method according to claim 1, characterized in that, The step of identifying the intent adjacent to the empty space intent in the second intent path and obtaining the first intent identification result of the empty space intent includes: In the second intent path, select the first adjacent intent and the second adjacent intent of the empty slot intent; Perform intent recognition on the first adjacent intent to obtain the second intent recognition result of the empty space intent; Intent recognition is performed on the second adjacent intent to obtain the third intent recognition result of the second adjacent intent; The first intent recognition result is determined based on the second intent recognition result and the third intent recognition result.

3. The method according to claim 2, characterized in that, The step of performing intent recognition on the first adjacent intent to obtain the second intent recognition result of the empty space intent includes: If the sentence type of the first adjacent intent is the first preset sentence type, then the target text is selected from the first adjacent intent, and the second intent recognition result includes the target text.

4. The method according to claim 2, characterized in that, The step of performing intent recognition on the first adjacent intent to obtain the second intent recognition result of the empty space intent includes: If the sentence type of the first adjacent intention is the second preset sentence type, then the first adjacent intention is input into the semantic feature recognition model for semantic feature recognition to obtain the semantic features in the first adjacent intention. The semantic features whose frequency of occurrence meets the preset frequency condition are selected as key semantic features, wherein the second intent recognition result includes the key semantic features.

5. The method according to claim 2, characterized in that, The step of performing intent recognition on the second adjacent intent to obtain a third intent recognition result for the second adjacent intent includes: The second adjacent intent is segmented into words to obtain the word segments of the second adjacent intent; By using the intent recognition model, feature extraction is performed on the word segmentation of the second adjacent intent to obtain word segmentation feature vector and word segmentation weight vector; The word segmentation feature vector and the word segmentation weight vector are identified to obtain the named entity of the second adjacent intent, wherein the third intent identification result includes the named entity.

6. The method according to claim 1, characterized in that, The step of generating a first intent path based on the first character's intent in the target scene includes: Obtain the intent of the first character in the target scene; The intentions of the first character in the target scene are sorted to obtain an intention sequence; The first intent path is determined based on the intent sequence.

7. The method according to claim 6, characterized in that, The step of sorting the intentions of the first character in the target scene to obtain an intention sequence includes: The intentions of the first character in the target scene are numbered in m rounds to obtain m sets of numbers, where m is a positive integer; Based on the intent numbers in the m number sets, the intents of the first character in the target scene are sorted to obtain m intent sequences; wherein, each intent sequence corresponds one-to-one with a second intent path; Determining the second character's intent in the target scene based on the first intent recognition result includes: Based on the first intent recognition results of the empty intents in the m second intent paths, the m intent groups of the second role in the target scene are determined.

8. The method according to claim 7, characterized in that, The method further includes: For each of the m intent groups, determine the first similarity between the intents in the intent group; In each intent group, intents whose first similarity reaches the similarity threshold are merged to obtain the intent merging result for each intent group; Based on the intent merging result of each intent group, the intent set of the second role in the target scene is determined.

9. The method according to claim 1, characterized in that, The number of intent recognition results is n; determining the intent of the second character in the target scene based on the first intent recognition result includes: Determine the second similarity among n intent recognition results; Among the n intent recognition results, the intent recognition results whose second similarity reaches the similarity threshold are merged to obtain the first merged intent recognition result; The intent of the second character in the target scene is determined based on the first merged intent recognition result.

10. The method according to claim 9, characterized in that, The method further includes: A third similarity score is obtained by calculating the similarity between the intentions adjacent to the empty space intention; Among the intentions adjacent to the empty space intention, the intentions whose third similarity reaches the similarity threshold are merged to obtain the second merged intention recognition result; Determining the intent of the second character in the target scene based on the first merged intent recognition result includes: Based on the first merged intent recognition result and the second merged intent recognition result, the intent of the second role in the target scene is determined.

11. An intent-determining device, characterized in that, The device includes: The generation module is used to generate a first intent path based on the first character's intent in the target scene; An insertion module is used to insert an empty intent into the first intent path to obtain a second intent path; The identification module is used to identify the intentions adjacent to the empty space intentions in the second intention path, and obtain the first intention identification result of the empty space intentions; The first determining module is used to determine the intention of the second character in the target scene based on the first intention recognition result.

12. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 10.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 10.

14. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 10.

Images