A response method and device

Abstract

The embodiment of the present invention provides a response method and device. The method includes: in the machine response link, obtaining the interactive dialogue that has occurred before the response time; determining the category to which each expression in the interactive dialogue belongs; forming the condition category of the interactive dialogue according to the dialogue order of each expression in the interactive dialogue; in each historical dialogue, determining the target category whose target value meets the set condition according to the condition category; and determining the response expression of the machine response link according to each historical dialogue that meets the target category whose target value meets the set condition. By obtaining at least one round of dialogue to determine the condition category, more consideration is given to the interactive logic between the machine and the user in this call, so that the determined response language is more reasonable and conforms to the interactive logic of this call. The response language determined by the above method is more reasonable, and because it is supported by practical data, no manual optimization is required, saving manpower time.

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of artificial intelligence technology, and in particular, to a response method, apparatus, computing device, and computer-readable storage medium. Background Art

[0002] Intelligent voice robots provide automatic outbound calls and natural responses, helping companies reduce labor costs and improve outbound call efficiency. In terms of implementation technology, intelligent voice robots integrate ASR (Automatic Speech Recognition), NLP (Natural Language Processing) and TTS (Text-To-Speech) modules. In the conversation process, the ASR module first converts the voice from the user end into text, the NLP module parses the text and predicts the response language, the TTS module converts the response language into voice, and finally the voice is transmitted to the user end through the line, forming a round of conversation interaction.

[0003] The logic of the NLP module is the most complex part of the voice robot, and its most widely used implementation method is the pipeline method. The pipeline method is generally divided into three steps, namely NLU (Natural Language Understanding), DM (Dialog Management) and NLG (Neural Language Generation). Among them, NLU understands the user text, and DM determines the category of the corresponding response language based on the understood user text according to a certain user and robot interaction strategy, thereby forming a complete dialogue logic. NLG determines the final output response language according to the category of the response language determined by DM.

[0004] In the prior art, DM designs interaction strategies based on the experience of industry experts. Experts will draw up dialogue flow charts based on their experience. This will cause the design of process branches to be strongly dependent on expert experience. Experts' experience-based enumeration will result in low coverage of response language. Moreover, the response language is highly subjective and is not supported by practical data. The rationality of the response language is questionable. At the same time, in order to ensure the response effect, it is necessary to manually read the dialogue history records, find out the response language with poor dialogue experience, and update and optimize it. It can be found that the optimization method is also very time-consuming and labor-intensive, and there is no systematicity.

[0005] In summary, the embodiment of the present invention provides a response method to improve the rationality of the response language and reduce the burden of manual optimization. Summary of the invention

[0006] The embodiment of the present invention provides a response method to improve the rationality of response language and reduce the burden of manual optimization.

[0007] In a first aspect, an embodiment of the present invention provides a response method, including:

[0008] In the machine answering phase, obtaining the interactive dialogue that has occurred before the answering moment; the interactive dialogue includes at least one round of dialogue between the machine and the user;

[0009] Determining the category to which each statement in the interactive dialogue belongs;

[0010] The categories to which the expressions belong are formed into condition categories of the interactive dialogue according to the dialogue order of the expressions in the interactive dialogue;

[0011] In each historical conversation, according to the condition category, a target category whose target value meets the set condition is determined, wherein the target category is the category to which the response expression located after the condition category in each historical conversation belongs;

[0012] The response expression of the machine response link is determined according to each historical conversation of the target category that meets the target value and satisfies the set conditions.

[0013] By extracting the condition category from at least one round of conversation, the optimal target category of the condition category is determined in each historical conversation, that is, the target category whose target value meets the set conditions. When determining the optimal target category, each historical conversation is introduced, so that the target category determined in this way is supported by big data, more reasonable, and has a positive significance for the transaction result of this call. At the same time, when determining the optimal target category, by determining the target value of each target category for the condition category, it is possible to more intuitively determine the optimal target category that has a positive effect on the condition category in each target category. In addition, by obtaining at least one round of conversation to determine the condition category, more consideration is given to the interaction logic between the machine and the user in this call, so that the determined response language is more reasonable and conforms to the interaction logic of this call. The response language determined by the above method is more reasonable, and because it is supported by practical data, no manual optimization is required, saving manpower time.

[0014] Optionally, before obtaining the interactive dialogue that has occurred before the answering moment, the method further includes:

[0015] For any historical conversation, determine the category to which each statement in the historical conversation belongs and the conversation result of the historical conversation, and integrate to form a category tree diagram of each historical conversation;

[0016] According to the condition category, the target category whose target value meets the set condition is determined in each historical dialogue, including:

[0017] In the category tree diagram, determining each target category corresponding to the condition category;

[0018] For any target category, determining a first dialogue result of a historical dialogue including the condition category and a second dialogue result of a historical dialogue including the condition category and the target category; determining a target value of the target category according to the first dialogue result and the second dialogue result;

[0019] Among the target categories, a target category whose target value satisfies the set conditions is determined.

[0020] By determining the target categories corresponding to the conditional categories in the category tree diagram obtained from the historical conversations, the conversations of each target category are analyzed to obtain the target values ​​of each target category, thereby determining the target category whose target value meets the set conditions. Thus, the optimal target category with a positive effect on the conditional category is determined in each target category. The determined response is more reasonable, and because it is supported by practical data, it does not require manual optimization, saving manpower time.

[0021] Optionally, before obtaining the interactive dialogue that has occurred before the answering moment, the method further includes:

[0022] For any historical conversation, determine the category to which each statement in the historical conversation belongs and the conversation result of the historical conversation, and integrate to form a category tree diagram of each historical conversation;

[0023] In the category tree diagram, each reference category and each corresponding target category that meets the set number of rounds are divided;

[0024] For any target category corresponding to any reference category, determine a third dialogue result of historical dialogues including the reference category and a fourth dialogue result of historical dialogues including the reference category and the target category; determine a target value of the target category corresponding to the reference category based on the third dialogue result and the fourth dialogue result;

[0025] For any reference category, among the target categories corresponding to the reference category, determine the target category whose target value meets the set conditions; store the corresponding relationship between the reference category and the target category whose target value meets the set conditions in a database;

[0026] According to the condition category, the target category whose target value meets the set condition is determined in each historical dialogue, including:

[0027] In the database, a reference category consistent with the condition category is determined, and according to the corresponding relationship between the reference category and the target category whose target value meets the set condition, the target category whose target value meets the set condition is determined.

[0028] The correspondence between each reference category and its corresponding target value that meets the set conditions is stored. Then, in the machine response link, after the condition category is determined, the target category whose target value corresponding to the condition category meets the set conditions can be directly determined in the correspondence. In this way, not only is the rationality of the determined response language guaranteed, but the speed of determining the response language is also accelerated. In addition, due to the support of practical data, there is no need for manual optimization, which saves manpower time.

[0029] Optionally, for any target category, determining a first dialogue result of a historical dialogue including the condition category and a second dialogue result of a historical dialogue including the condition category and the target category includes:

[0030] In the historical conversations including the condition category, the ratio of the number of historical conversations whose conversation results meet the first preset threshold to the number of historical conversations including the condition category is determined as the first conversation result;

[0031] For any target category, in the historical conversations including the condition category and the target category, the ratio of the number of historical conversations whose conversation results meet the first preset threshold to the number of historical conversations including the condition category and the target category is determined as the second conversation result;

[0032] Determining a target value of the target category according to the first conversation result and the second conversation result includes:

[0033] A ratio of the first dialogue result to the second dialogue result is determined as a target value of the target category.

[0034] The historical conversations containing the conditional category are analyzed to determine the first conversation result, and the historical conversations containing the conditional category and the target category are analyzed to determine the second conversation result. In this way, it can be determined whether any target category has a positive or negative effect on the conditional category. It is more intuitive and reasonable to quantify this effect in the form of a target value.

[0035] Optionally, determining a target category whose target value meets a set condition in each target category includes:

[0036] Among the target categories, the target category having a target value greater than the second preset threshold and the largest target category is determined as the target category whose target value satisfies the set condition.

[0037] The target value refers to the effect of any target category on the condition category. Only when it meets the second preset threshold can it meet the basic requirements of the developer for the positive effect on the target category. Determining the target category with the largest target value among multiple target categories is conducive to selecting the optimal target category, thereby determining the most reasonable response.

[0038] Optionally, the category to which each of the expressions belongs is determined by the following method, including:

[0039] For any expression, vectorize the expression to obtain a vectorized expression of the expression;

[0040] For any vectorized expression to be determined, determine its similarity with the existing vectors of each first category; if the similarity with any vector of the first category meets the third preset threshold, classify the vectorized expression to be determined into the first category and update the vector of the first category; if the similarity with any first category does not meet the third preset threshold, establish a second category for the vectorized expression to be determined. The vectors of the first category are obtained from the vectorized expressions in the first category; the vectorized expression to be determined is a vectorized expression whose category is not determined.

[0041] Determining the category of any statement using the above method is conducive to subsequent simplification of operations and enhancing the rationality of the determined response.

[0042] Optionally, vectorizing the expression to obtain a vectorized expression of the expression includes:

[0043] Performing word segmentation processing on the expression, extracting text features from the expression after word segmentation processing, and obtaining a text vectorized expression;

[0044] Extracting semantic features from the expression after word segmentation processing to obtain a semantic vectorized expression;

[0045] The text vectorized expression and the semantic vectorized expression are concatenated to obtain a vectorized expression of the expression.

[0046] By extracting text features and semantic features from any expression, the resulting vectorized expression comprehensively considers literal information and semantic information, has richer levels and more comprehensive information, and is conducive to improving the accuracy of determining the category to which the expression belongs.

[0047] In a second aspect, an embodiment of the present invention further provides a response device, including:

[0048] An acquisition unit, used to acquire, during the machine answering phase, an interactive dialogue that has occurred before the answering moment; the interactive dialogue includes at least one round of dialogue between the machine and the user;

[0049] Identify units for:

[0050] Determining the category to which each statement in the interactive dialogue belongs;

[0051] The categories to which the expressions belong are formed into condition categories of the interactive dialogue according to the dialogue order of the expressions in the interactive dialogue;

[0052] According to the condition category, determining a target category whose target value meets the set condition in each historical conversation, the target category being the category to which the response expression located after the condition category in each historical conversation belongs;

[0053] The response expression of the machine response link is determined according to each historical dialogue of the target category whose target value meets the set conditions.

[0054] Optionally, the determining unit is further configured to:

[0055] For any historical conversation, determine the category to which each statement in the historical conversation belongs and the conversation result of the historical conversation, and integrate to form a category tree diagram of each historical conversation;

[0056] The determining unit is specifically configured to:

[0057] In the category tree diagram, determining each target category corresponding to the condition category;

[0058] For any target category, determining a first dialogue result of a historical dialogue including the condition category and a second dialogue result of a historical dialogue including the condition category and the target category; determining a target value of the target category according to the first dialogue result and the second dialogue result;

[0059] Among the target categories, a target category whose target value satisfies the set conditions is determined.

[0060] Optionally, the determining unit is further configured to:

[0061] For any historical conversation, determine the category to which each statement in the historical conversation belongs and the conversation result of the historical conversation, and integrate to form a category tree diagram of each historical conversation;

[0062] In the category tree diagram, each reference category and each corresponding target category that meets the set number of rounds are divided;

[0063] For any target category corresponding to any reference category, determine a third dialogue result of historical dialogues including the reference category and a fourth dialogue result of historical dialogues including the reference category and the target category; determine a target value of the target category corresponding to the reference category based on the third dialogue result and the fourth dialogue result;

[0064] For any reference category, among the target categories corresponding to the reference category, determine the target category whose target value meets the set conditions; store the corresponding relationship between the reference category and the target category whose target value meets the set conditions in a database;

[0065] The determining unit is specifically configured to:

[0066] In the database, a reference category consistent with the condition category is determined, and according to the corresponding relationship between the reference category and the target category whose target value meets the set condition, the target category whose target value meets the set condition is determined.

[0067] Optionally, the determining unit is specifically configured to:

[0068] In the historical conversations including the condition category, the ratio of the number of historical conversations whose conversation results meet the first preset threshold to the number of historical conversations including the condition category is determined as the first conversation result;

[0069] For any target category, in the historical conversations including the condition category and the target category, the ratio of the number of historical conversations whose conversation results meet the first preset threshold to the number of historical conversations including the condition category and the target category is determined as the second conversation result;

[0070] Determining a target value of the target category according to the first conversation result and the second conversation result includes:

[0071] A ratio of the first dialogue result to the second dialogue result is determined as a target value of the target category.

[0072] Optionally, the determining unit is specifically configured to:

[0073] Among the target categories, the target category having a target value greater than the second preset threshold and the largest target category is determined as the target category whose target value satisfies the set condition.

[0074] Optionally, the determining unit is specifically configured to:

[0075] For any expression, vectorize the expression to obtain a vectorized expression of the expression;

[0076] For any vectorized expression to be determined, determine its similarity with the existing vectors of each first category; if the similarity with any vector of the first category meets the third preset threshold, classify the vectorized expression to be determined into the first category and update the vector of the first category; if the similarity with any first category does not meet the third preset threshold, establish a second category for the vectorized expression to be determined. The vectors of the first category are obtained from the vectorized expressions in the first category; the vectorized expression to be determined is a vectorized expression whose category is not determined.

[0077] Optionally, the determining unit is specifically configured to:

[0078] Performing word segmentation processing on the expression, extracting text features from the expression after word segmentation processing, and obtaining a text vectorized expression;

[0079] Extracting semantic features from the expression after word segmentation processing to obtain a semantic vectorized expression;

[0080] The text vectorized expression and the semantic vectorized expression are concatenated to obtain a vectorized expression of the expression.

[0081] In a third aspect, an embodiment of the present invention further provides a computing device, including:

[0082] Memory for storing computer programs;

[0083] The processor is used to call the computer program stored in the memory and execute the response method listed in any of the above-mentioned ways according to the obtained program.

[0084] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer-executable program, and the computer-executable program is used to enable a computer to execute any of the response methods listed in the above manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0086] Figure 1 A schematic diagram of a system architecture provided for an embodiment of the present invention;

[0087] Figure 2 A schematic diagram of a response method provided by an embodiment of the present invention;

[0088] Figure 3 A schematic diagram of a category tree diagram provided by an embodiment of the present invention;

[0089] Figure 4 A schematic diagram of a category tree diagram provided by an embodiment of the present invention;

[0090] Figure 5 A schematic diagram of a response method provided by an embodiment of the present invention;

[0091] Figure 6 A schematic diagram of a response method provided by an embodiment of the present invention;

[0092] Figure 7A schematic diagram of the structure of a response device provided by an embodiment of the present invention;

[0093] Figure 8 A schematic diagram of the structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0094] In order to make the purpose, implementation mode and advantages of the present application clearer, the exemplary implementation mode of the present application will be clearly and completely described below in conjunction with the drawings in the exemplary embodiments of the present application. Obviously, the described exemplary embodiments are only part of the embodiments of the present application, rather than all the embodiments.

[0095] Based on the exemplary embodiments described in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the claims attached to this application. In addition, although the disclosure in this application is introduced according to one or several exemplary examples, it should be understood that each aspect of the disclosure can also constitute a complete implementation method separately.

[0096] It should be noted that the brief description of terms in this application is only for the convenience of understanding the embodiments described below, and is not intended to limit the embodiments of this application. Unless otherwise specified, these terms should be understood according to their common and usual meanings.

[0097] The terms "first", "second", "third", etc. in the specification and claims of this application and the above drawings are used to distinguish similar or similar objects or entities, and do not necessarily mean to limit a specific order or sequence, unless otherwise indicated. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances, for example, they can be implemented according to orders other than those given in the diagrams or descriptions of the embodiments of this application.

[0098] In addition, the terms "include" and "have" and any variations thereof are intended to cover but not exclude inclusion, for example, a product or device comprising a list of components is not necessarily limited to those components expressly listed but may include other components not expressly listed or inherent to such products or devices.

[0099] Figure 1 A system architecture applicable to an embodiment of the present invention is exemplarily shown. The system architecture may be a server 100 , including a processor 110 , a communication interface 120 , and a memory 130 .

[0100] The communication interface 120 is used to communicate with the terminal device, send and receive information transmitted by the terminal device, and realize communication.

[0101] The processor 110 is the control center of the server 100, and uses various interfaces and routes to connect various parts of the entire server 100, and executes various functions of the server 100 and processes data by running or executing software programs and / or modules stored in the memory 130, and calling data stored in the memory 130. Optionally, the processor 110 may include one or more processing units.

[0102] The memory 130 can be used to store software programs and modules. The processor 110 executes various functional applications and data processing by running the software programs and modules stored in the memory 130. The memory 130 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function, etc.; the data storage area can store data created according to business processing, etc. In addition, the memory 130 can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0103] It should be noted that the above Figure 1 The structure shown is only an example and is not limited to this embodiment of the present invention.

[0104] Above Figure 1 The server shown can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0105] The embodiment of the present invention provides a response method, such as Figure 2 As shown, including:

[0106] Step 201: In the machine answering phase, obtain the interactive dialogue that has occurred before the answering moment; the interactive dialogue includes at least one round of dialogue between the machine and the user;

[0107] Step 202: determining the category to which each statement in the interactive dialogue belongs;

[0108] Step 203: The categories to which the expressions belong are grouped according to the order of the expressions in the interactive dialogue to form condition categories of the interactive dialogue;

[0109] Step 204: in each historical conversation, according to the condition category, determine a target category whose target value meets the set condition, wherein the target category is the category to which the response expression located after the condition category in each historical conversation belongs;

[0110] Step 205: Determine the response expression of the machine response link according to each historical conversation of the target category that meets the target value and satisfies the set conditions.

[0111] Collect real conversations between excellent agents and users, or between machines and users, form historical conversations, and mark the conversation results of each historical conversation. The conversation results can be whether the transaction is completed, user satisfaction, etc. For example, a conversation is a conversation between an agent and a user. Within a set time (such as one day) after the end of the call, the user pays off the account, then the conversation result of the historical conversation is 1; if the user does not pay off the account within the set time, then the conversation result of the historical conversation is 0. For example, after any conversation ends, collect the user's satisfaction. If the user chooses "satisfied", the conversation result of the historical conversation is 1; if the user chooses "unsatisfied", the conversation result of the historical conversation is 0. The above is only an example. The embodiment of the present invention does not limit the content of the conversation result. It only needs to collect the conversation result of each historical conversation to guide the determination of the answer expression of the machine answering link.

[0112] The following describes a method for determining the category to which each statement in each collected historical conversation belongs.

[0113] Step 1: Use the word segmentation tool to segment the expression, and then vectorize the expression to obtain the vectorized expression.

[0114] There are many methods for vectorization, such as literal method, semantic method, etc.

[0115] The literal method mainly selects fine-grained features of the text, for example, you can use term frequency-inverse document frequency (TF-IDF) as a feature, and add n-gram information to consider word order. The vector dimension obtained by the above steps is large, and the principal component analysis method (PCA) can be used to reduce the dimension. For example, after the literal method is processed, the expression: I am / happy / to see / you will get a literal quantitative expression (1,2,3,4).

[0116] Semantic methods can calculate the semantic features of expressions, such as using the Word2vec method to train word vectors to obtain the semantic representation of each word after considering the context, and using the unsupervised SIF method to weighted average the word vectors to obtain the semantic vector of the expression. For example, after being processed by the semantic method, the expression: I am / happy / to see / you will obtain a semantic vectorized expression (5,6,7,8).

[0117] The above two methods can be used alone or in combination, depending on the usage scenario. If the accuracy of the expression conversion is not required, and the expressions in the historical dialogue are relatively simple, then only the literal method can be used. If the accuracy of the expression is required to be high, and the expressions in the historical dialogue are relatively rich, the two methods can be combined to splice the obtained literal quantitative expression and semantic vectorized expression to obtain a vectorized expression that comprehensively considers the literal information and semantic information. For example (1,2,3,4,5,6,7,8). The above is only an example.

[0118] Step 2: for any vectorized expression to be determined, determine its similarity with the existing vectors of each first category; if the similarity with any vector of the first category meets the third preset threshold, classify the vectorized expression to be determined into the first category and update the vector of the first category; if the similarity with any first category does not meet the third preset threshold, establish a second category for the vectorized expression to be determined. The vectors of the first category are obtained from the vectorized expressions in the first category.

[0119] Taking the clustering of user expressions in historical conversations as an example, select any vectorized expression to be determined, obtain the first category U0 (U0 contains the vectorized expression), and take the average value of each vectorized expression in the first category as the vector of the first category. 0 Construct an approximate nearest neighbor search tree T 0 . Traverse each vectorized representation u to be determined in turn i , determine u i The similarity between u and the vector of the first category is calculated. If the similarity meets the third preset threshold, u i The vector of the first category is updated. If the similarity does not meet the third preset threshold, then u i Create a second category and perform the above operations on each vectorized expression to divide each vectorized expression into multiple categories.

[0120] The second clustering method is introduced below, that is, the vectors of each clustering category are not calculated. Select any vectorized expression to be determined to obtain the first category U0, and then select any vectorized expression to be determined to determine the similarity between the vectorized expression to be determined and the vectorized expression in the first category. If the similarity meets the third preset threshold, the vectorized expression to be determined is divided into the first category U0. If it does not meet the third preset threshold, the vectorized expression to be determined is newly created as the second category, and the above process is repeated. When the first category or the second category contains more than two vectorized expressions, for any vectorized expression to be determined, the category to which it belongs is determined by voting. For example, there are 10 vectorized expressions in the first category and 3 vectorized expressions in the second category. For any vectorized expression to be determined, its similarity with the 10 vectorized expressions in the first category and the 3 vectorized expressions in the second category is determined respectively, and among the 13 similarity values, the similarity value that meets the third preset threshold is determined. If 3 are determined, of which 2 similarity values ​​correspond to the first category and 1 similarity value corresponds to the second category, then the vectorized expression to be determined is classified into the first category; alternatively, the category of the vectorized expression corresponding to the largest value among the 13 similarity values ​​is taken as the category to which the vectorized expression to be determined belongs.

[0121] The above is only an example, and the embodiment of the present invention does not limit the clustering method.

[0122] By using the above method, each user expression in the historical conversation is clustered to obtain cluster categories U0, U1, U2, ...; each machine expression in the historical conversation is clustered to obtain cluster categories R0, R1, R2, ...;

[0123] Then for each historical conversation, each expression can be replaced with the corresponding category in the order of the conversation. For example, r represents machine expression and u represents user expression. A historical conversation is: r0-u0-r1-u1-r2-u2, where r0 belongs to R0, u0 belongs to U0, r1 belongs to R2, u1 belongs to U0, r2 belongs to R1, and u2 belongs to U4. Then the historical conversation is converted into: R0-U0-R2-U0-R1-U4.

[0124] Integrate various historical dialogues to form a category tree diagram of each historical dialogue, such as Figure 3 As shown. Figure 3 In the example, the first sentence of each historical dialogue is a machine statement, and the first sentence belongs to the same category R0. For the machine statement R0, the user gives a variety of statements of different categories, such as U1, U2, etc. Figure 3 This is just an example. Each historical conversation may be integrated into multiple category tree diagrams, which is not limited in this embodiment of the present invention.

[0125] exist Figure 3 The dialog results of various historical dialogs are also shown. If the same type of dialogs are not merged in the process of integrating into the category tree diagram, the dialog result still maintains the dialog result of each historical dialog (such as 0, 1); if the same type of dialogs are merged in the process of integrating into the category tree diagram, the dialog result of each historical dialog may change. For example, two historical dialogs belong to the same category of historical dialogs, and the dialog results of the two historical dialogs are 0 and 1 respectively. Then the dialog result of the historical dialog of this category is averaged, which is 0.5. The above is only an example, and the embodiment of the present invention is not limited to this.

[0126] against Figure 3 The category tree diagram shown in the figure can be divided into conditional categories and target categories. The conditional category can be one or more rounds of dialogue, and the target category is the category to which the response expression made by the machine for the conditional category belongs. Figure 4 As shown in the figure, if one round of dialogue is used as the condition category, the condition category can be R0-U1, R0-U2, R1-U3, etc. If two rounds of dialogue are used as the condition category, the condition category can be R0-U1-R1-U3. If the condition category is R0-U1, the corresponding target category is R1 and R2; if the condition category is R0-U2, the corresponding target category is R1; if the condition category is R0-U1-R1-U3, the corresponding target category is R2. Since the user dialogues contained in the historical dialogues are very rich, each condition category may correspond to more than one target category. Therefore, it is necessary to make trade-offs among the target categories and select the optimal target category to guide the determination of the machine's response expression in the machine response link.

[0127] One possible implementation method is to determine the categories of each expression in the interactive dialogue that has occurred before the answering moment in the machine answering phase, form conditional categories, and Figure 3 In the category tree diagram shown, multiple target categories of the machine corresponding to the condition category are determined, and the optimal target category is selected from the multiple target categories. This method first determines the condition category, and then specifically determines the optimal target category of the condition category in the category tree diagram. This method does not require a lot of work in the early stage, the process is simple, and the flexibility is strong. The number of rounds of interactive dialogue can be changed at will. Another possible implementation method is to first target Figure 3 The category tree diagram shown in the figure determines the multiple target categories corresponding to various reference categories (which are consistent with the concept of conditional categories. In order to distinguish, the conditional categories determined in advance in the historical dialogue are named reference categories). The optimal target category is selected from the multiple target categories. The corresponding relationship between the reference category and the optimal target category is stored in the database. This requires considering all possible situations. For example, if the number of dialogue rounds is set to one, then in Figure 3In the category tree diagram shown, all possible reference categories for one round are determined, and for each reference category, the optimal target category is selected for storage. Then, in the subsequent machine response phase, after the conditional category is determined, the optimal target category corresponding to the reference category consistent with the conditional category can be directly determined in the database. This method requires a lot of work in the early stage and is cumbersome, but it reduces the workload in the machine response phase and can make response statements faster. It is more real-time.

[0128] The following is a detailed introduction to the two methods mentioned above.

[0129] Method 1: Figure 5 As shown, the following steps are included:

[0130] Step 501: In the machine answering phase, obtain the interactive dialogue that has occurred before the answering moment.

[0131] The interactive dialogue includes at least one round of dialogue between the machine and the user. For example, r represents the machine expression and u represents the user expression. Then the obtained interactive dialogue can be: r0-u0 (one round of dialogue), r0-u0-r1-u1 (two rounds of dialogue), r0-u0-r1-u1-r2-u2 (three rounds of dialogue), and so on.

[0132] It can be found that the more rounds of dialogue obtained, the more the previous dialogue logic between the machine and the user can be taken into account, ensuring that the subsequent machine response expression will not be repeated with the previous machine expression, and the rationality of the subsequent machine response expression can be guaranteed. However, if the number of rounds is too large, the number of target categories determined in the category tree diagram will be small, the data will be sparse, and the accuracy may be affected. Therefore, the number of rounds of dialogue is generally selected to be two or three rounds.

[0133] Step 502: Determine the category to which each expression in the interactive dialogue belongs.

[0134] The expressions in the interactive dialogue are clustered according to the method for clustering historical dialogues introduced above, and the initial categories of clustering are the categories obtained after clustering the historical dialogues; or, the initial categories may not be set, and the expressions in the interactive dialogue are clustered again. The embodiment of the present invention is not limited to this.

[0135] Step 503: The categories to which the expressions belong are combined according to the dialog order of the expressions in the interactive dialog to form condition categories of the interactive dialog.

[0136] For example, the resulting condition category is R0-U1.

[0137] Step 504: In the category tree diagram, determine each target category corresponding to the condition category.

[0138] Traverse as Figure 3 The category tree diagram shown determines the target categories corresponding to the condition categories R0-U1, for example, the determined target categories are R1 and R2.

[0139] Step 505: For any target category, determine a first dialogue result of a historical dialogue including the condition category and a second dialogue result of a historical dialogue including the condition category and any target category.

[0140] In historical conversations that include the condition category, the first conversation result is determined as the ratio of the number of historical conversations whose conversation results meet the first preset threshold to the number of historical conversations that include the condition category; for any target category, in historical conversations that include the condition category and any target category, the second conversation result is determined as the ratio of the number of historical conversations whose conversation results meet the first preset threshold to the number of historical conversations that include the condition category and any target category.

[0141] For example, in Figure 3 In the shown category tree diagram, there are 3 calls with condition category R0-U1, of which 2 calls have a conversation result of 1, so the first conversation result is 2 / 3; there are 2 calls with condition category R0-U1 and target category R1, of which 2 calls have a conversation result of 1, so the second conversation result with target category R1 is 1. There is 1 call with condition category R0-U1 and target category R2, of which 0 calls have a conversation result of 1, so the second conversation result with target category R2 is 0.

[0142] Step 506: Determine a target value of any target category according to the first dialogue result and the second dialogue result.

[0143] A ratio of the first dialogue result to the second dialogue result is determined as a target value of any target category.

[0144] For example, in the example of step 505, the target value of the target category R1 is 3 / 2, and the target value of the target category R2 is 0. The target value is greater than 1, indicating that the target category has a positive promoting effect on the historical dialogues belonging to the condition category. In the above example, when the historical dialogue between the machine and the user is only the condition category (R0-U1), the success rate of the historical dialogue (the first dialogue result) is 2 / 3, but when the machine responds to the target category (R1) according to the condition category, the success rate of the historical dialogue (the second dialogue result) is 1. The target value is 3 / 2, so it can be seen that R1 has a positive promoting effect on the historical dialogues belonging to the condition category R0-U1.

[0145] Step 507: Determine, among the target categories, the target category whose target value meets the set conditions.

[0146] The setting condition can be that the target value is greater than the second preset threshold, or the target value is the maximum, or the target value is greater than the second preset threshold and the target value is the maximum. Assume that in the example of step 505, target values ​​of three target categories are obtained, namely 3 / 2, 4 / 5, and 7 / 3. Select the target category corresponding to the target value greater than 1 and the maximum target value as the target category whose target value meets the setting condition. Then in the above example, the target category corresponding to 7 / 3 is selected as the target category whose target value meets the setting condition. The second preset threshold can also be selected as 1.5, 2, etc., and those skilled in the art can choose according to actual needs.

[0147] Method 2: Figure 6 As shown, the following steps are included:

[0148] Step 601: for any historical conversation, determine the category to which each statement in the historical conversation belongs and the conversation result of the historical conversation, and integrate them to form a category tree diagram of each historical conversation. Figure 3 shown.

[0149] Step 602: In the category tree diagram, determine multiple target categories corresponding to various reference categories.

[0150] If the number of rounds is set to one, then in the category tree diagram, multiple target categories corresponding to various reference categories with only one round of dialogue are determined, for example, the multiple target categories corresponding to various reference categories are determined as shown in Table 1. If the number of rounds is set to two, then in the category tree diagram, multiple target categories corresponding to various reference categories with only two rounds of dialogue are determined.

[0151] Table 1

[0152]

[0153] Step 603: For any reference category, select the best target category from multiple target categories, that is, the target category whose target value meets the set conditions. The determination method is the same as steps 505, 506 and 507 in method 1, and will not be repeated here.

[0154] Step 604: Store the corresponding relationship between any reference category and the optimal target category in a database. Table 2 schematically shows a possible storage format.

[0155] Table 2

[0156]

[0157]

[0158] Step 605: In the machine answering phase, the interactive dialogue that has occurred before the answering moment is obtained.

[0159] Step 606: Determine the category to which each statement in the interactive dialogue belongs.

[0160] Step 607: The categories to which the expressions belong are grouped according to the dialog order of the expressions in the interactive dialog to form condition categories of the interactive dialog.

[0161] Step 608: Determine in the database the optimal target category corresponding to the reference category consistent with the condition category. For example, if the condition category determined in step 607 is R0-U1, then the optimal target category is directly determined in the database as R1.

[0162] In the introduction Figure 6 In the method, individual process methods are the same Figure 5 The process method is similar to that of Figure 5 I will not go into details here.

[0163] After determining the target category whose target value meets the set conditions, the response expression can be determined according to the target category. Specifically, in the historical conversation, the machine expression belonging to the target category is selected as the response expression of the current machine response link. For example, in the previously determined cluster categories, the category with the target category R1 contains machine expressions r1, r2 and r3. Then, one machine expression is selected from r1, r2 and r3 as the response expression of the current machine response link for output.

[0164] Optionally, in the machine answering phase, the number of rounds of dialogue between the machine and the user in the obtained interactive dialogue can be adjusted flexibly. For example, first obtain the three rounds of dialogue that have occurred before the answering moment, and if the optimal target category can be determined in each historical dialogue, select any machine expression in the target category for output; if the optimal target category cannot be determined in each historical dialogue, reduce the number of rounds, obtain the two rounds of dialogue that have occurred before the answering moment, and repeat the previous steps until the optimal target category can be determined.

[0165] Based on the same technical concept, Figure 7 The structure of a response device provided by an embodiment of the present invention is exemplarily shown, and the structure can execute the response process.

[0166] like Figure 7 As shown, the device specifically includes:

[0167] The acquisition unit 701 is used to acquire the interactive dialogue that has occurred before the answering moment during the machine answering phase; the interactive dialogue includes at least one round of dialogue between the machine and the user;

[0168] The determining unit 702 is configured to:

[0169] Determining the category to which each statement in the interactive dialogue belongs;

[0170] The categories to which the expressions belong are formed into condition categories of the interactive dialogue according to the dialogue order of the expressions in the interactive dialogue;

[0171] According to the condition category, determining a target category whose target value meets the set condition in each historical conversation, the target category being the category to which the response expression located after the condition category in each historical conversation belongs;

[0172] The response expression of the machine response link is determined according to each historical dialogue of the target category whose target value meets the set conditions.

[0173] Optionally, the determining unit 702 is further configured to:

[0174] For any historical conversation, determine the category to which each statement in the historical conversation belongs and the conversation result of the historical conversation, and integrate to form a category tree diagram of each historical conversation;

[0175] The determining unit 702 is specifically configured to:

[0176] In the category tree diagram, determining each target category corresponding to the condition category;

[0177] For any target category, determining a first dialogue result of a historical dialogue including the condition category and a second dialogue result of a historical dialogue including the condition category and the target category; determining a target value of the target category according to the first dialogue result and the second dialogue result;

[0178] Among the target categories, a target category whose target value satisfies the set conditions is determined.

[0179] Optionally, the determining unit 702 is further configured to:

[0180] For any historical conversation, determine the category to which each statement in the historical conversation belongs and the conversation result of the historical conversation, and integrate to form a category tree diagram of each historical conversation;

[0181] In the category tree diagram, each reference category and each corresponding target category that meets the set number of rounds are divided;

[0182] For any target category corresponding to any reference category, determine a third dialogue result of historical dialogues including the reference category and a fourth dialogue result of historical dialogues including the reference category and the target category; determine a target value of the target category corresponding to the reference category based on the third dialogue result and the fourth dialogue result;

[0183] For any reference category, among the target categories corresponding to the reference category, determine the target category whose target value meets the set conditions; store the corresponding relationship between the reference category and the target category whose target value meets the set conditions in a database;

[0184] The determining unit 702 is specifically configured to:

[0185] In the database, a reference category consistent with the condition category is determined, and according to the corresponding relationship between the reference category and the target category whose target value meets the set condition, the target category whose target value meets the set condition is determined.

[0186] Optionally, the determining unit 702 is specifically configured to:

[0187] In the historical conversations including the condition category, the ratio of the number of historical conversations whose conversation results meet the first preset threshold to the number of historical conversations including the condition category is determined as the first conversation result;

[0188] For any target category, in the historical conversations including the condition category and the target category, the ratio of the number of historical conversations whose conversation results meet the first preset threshold to the number of historical conversations including the condition category and the target category is determined as the second conversation result;

[0189] The determining unit 702 is specifically configured to:

[0190] A ratio of the first dialogue result to the second dialogue result is determined as a target value of the target category.

[0191] Optionally, the determining unit 702 is specifically configured to:

[0192] Among the target categories, the target category having a target value greater than the second preset threshold and the largest target category is determined as the target category whose target value satisfies the set condition.

[0193] Optionally, the determining unit 702 is specifically configured to:

[0194] For any expression, vectorize the expression to obtain a vectorized expression of the expression;

[0195] For any vectorized expression to be determined, determine its similarity with the existing vectors of each first category; if the similarity with any vector of the first category meets the third preset threshold, classify the vectorized expression to be determined into the first category and update the vector of the first category; if the similarity with any first category does not meet the third preset threshold, establish a second category for the vectorized expression to be determined. The vectors of the first category are obtained from the vectorized expressions in the first category; the vectorized expression to be determined is a vectorized expression whose category is not determined.

[0196] Optionally, the determining unit 702 is specifically configured to:

[0197] Performing word segmentation processing on the expression, extracting text features from the expression after word segmentation processing, and obtaining a text vectorized expression;

[0198] Extracting semantic features from the expression after word segmentation processing to obtain a semantic vectorized expression;

[0199] The text vectorized expression and the semantic vectorized expression are concatenated to obtain a vectorized expression of the expression.

[0200] Based on the same technical concept, the embodiment of the present application provides a computer device, such as Figure 8 As shown, it includes at least one processor 801 and a memory 802 connected to the at least one processor. The specific connection medium between the processor 801 and the memory 802 is not limited in the embodiment of the present application. Figure 8 For example, the processor 801 and the memory 802 are connected via a bus. The bus can be divided into an address bus, a data bus, a control bus, etc.

[0201] In the embodiment of the present application, the memory 802 stores instructions that can be executed by at least one processor 801, and the at least one processor 801 can perform the steps of the above-mentioned response method by executing the instructions stored in the memory 802.

[0202] The processor 801 is the control center of the computer device, and can use various interfaces and lines to connect various parts of the computer device, and respond by running or executing instructions stored in the memory 802 and calling data stored in the memory 802. Optionally, the processor 801 may include one or more processing units, and the processor 801 may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface, and application programs, and the modem processor mainly processes wireless communications. It is understandable that the above-mentioned modem processor may not be integrated into the processor 801. In some embodiments, the processor 801 and the memory 802 may be implemented on the same chip, and in some embodiments, they may also be implemented separately on independent chips.

[0203] Processor 801 can be a general-purpose processor, such as a central processing unit (CPU), a digital signal processor, an application-specific integrated circuit (ASIC), a field programmable gate array or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, and can implement or execute the methods, steps and logic block diagrams disclosed in the embodiments of the present application. A general-purpose processor can be a microprocessor or any conventional processor, etc. The steps of the method disclosed in the embodiments of the present application can be directly embodied as a hardware processor for execution, or can be executed by a combination of hardware and software modules in the processor.

[0204] The memory 802 is a non-volatile computer-readable storage medium that can be used to store non-volatile software programs, non-volatile computer executable programs and modules. The memory 802 may include at least one type of storage medium, such as a flash memory, a hard disk, a multimedia card, a card-type memory, a random access memory (Random Access Memory, RAM), a static random access memory (Static Random Access Memory, SRAM), a programmable read-only memory (Programmable Read Only Memory, PROM), a read-only memory (Read Only Memory, ROM), an electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, EEPROM), a magnetic memory, a disk, an optical disk, etc. The memory 802 is any other medium that can be used to carry or store a desired program code in the form of an instruction or data structure and can be accessed by a computer, but is not limited thereto. The memory 802 in the embodiment of the present application can also be a circuit or any other device that can realize a storage function, for storing program instructions and / or data.

[0205] Based on the same technical concept, an embodiment of the present invention further provides a computer-readable storage medium, which stores a computer-executable program, and the computer-executable program is used to enable a computer to execute the response method listed in any of the above-mentioned ways.

[0206] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0207] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0208] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0209] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0210] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A response method, characterized in that: include: For any historical conversation, determine the category to which each statement in the historical conversation belongs and the conversation result of the historical conversation, and integrate to form a category tree diagram of each historical conversation; In the machine answering phase, obtaining the interactive dialogue that has occurred before the answering moment; the interactive dialogue includes at least one round of dialogue between the machine and the user; Determining the category to which each statement in the interactive dialogue belongs; The categories to which the expressions belong are formed into condition categories of the interactive dialogue according to the dialogue order of the expressions in the interactive dialogue; According to the condition category, determining a target category whose target value meets the set condition in each historical conversation, the target category being the category to which the response expression located after the condition category in each historical conversation belongs; Determining a response statement of the machine response link according to each historical conversation of the target category whose target value meets the set conditions; Determining the target category whose target value meets the set condition in each historical dialogue according to the condition category includes: In the category tree diagram, each target category corresponding to the condition category is determined; in historical conversations containing the condition category, the ratio of the number of historical conversations whose conversation results meet the first preset threshold to the number of historical conversations containing the condition category is determined as a first conversation result; for any target category, in historical conversations containing the condition category and the target category, the ratio of the number of historical conversations whose conversation results meet the first preset threshold to the number of historical conversations containing the condition category and the target category is determined as a second conversation result; the ratio of the first conversation result to the second conversation result is determined as the target value of the target category; and in each target category, a target category whose target value meets the set conditions is determined.

2. The method according to claim 1, characterized in that Before obtaining the interactive dialogue that has occurred before the answer time, it also includes: For any historical conversation, determine the category to which each statement in the historical conversation belongs and the conversation result of the historical conversation, and integrate to form a category tree diagram of each historical conversation; In the category tree diagram, each reference category and each corresponding target category that meets the set number of rounds are divided; For any target category corresponding to any reference category, determine a third dialogue result of historical dialogues including the reference category and a fourth dialogue result of historical dialogues including the reference category and the target category; determine a target value of the target category corresponding to the reference category based on the third dialogue result and the fourth dialogue result; For any reference category, among the target categories corresponding to the reference category, determine the target category whose target value meets the set conditions; store the corresponding relationship between the reference category and the target category whose target value meets the set conditions in a database; According to the condition category, the target category whose target value meets the set condition is determined in each historical dialogue, including: In the database, a reference category consistent with the condition category is determined, and according to the corresponding relationship between the reference category and the target category whose target value meets the set condition, the target category whose target value meets the set condition is determined.

3. The method according to claim 1, characterized in that In each target category, the target category whose target value meets the set conditions is determined, including: Among the target categories, the target category having a target value greater than the second preset threshold and the largest target category is determined as the target category whose target value satisfies the set condition.

4. The method according to any one of claims 1 to 3, characterized in that: The categories to which the expressions belong are determined by the following method, including: For any expression, vectorize the expression to obtain a vectorized expression of the expression; For any vectorized expression to be determined, determine its similarity with the existing vectors of each first category; if the similarity with any vector of the first category meets the third preset threshold, classify the vectorized expression to be determined into the first category, and update the vector of the first category; if the similarity with any first category does not meet the third preset threshold, establish a second category for the vectorized expression to be determined; the vector of the first category is obtained from the vectorized expressions in the first category; the vectorized expression to be determined is a vectorized expression whose category is not determined.

5. The method according to claim 4, characterized in that The expression is vectorized to obtain a vectorized expression of the expression, including: Performing word segmentation processing on the expression, extracting text features from the expression after word segmentation processing, and obtaining a text vectorized expression; Extracting semantic features from the expression after word segmentation processing to obtain a semantic vectorized expression; The text vectorized expression and the semantic vectorized expression are concatenated to obtain a vectorized expression of the expression.

6. A response device, characterized in that: include: An acquisition unit is used to determine, for any historical conversation, the category to which each statement in the historical conversation belongs and the conversation result of the historical conversation, integrate and form a category tree diagram of each historical conversation, and in the machine answering stage, acquire the interactive conversation that has occurred before the answering moment; the interactive conversation includes at least one round of conversation between the machine and the user; Identify units for: Determining the category to which each statement in the interactive dialogue belongs; The categories to which the expressions belong are formed into condition categories of the interactive dialogue according to the dialogue order of the expressions in the interactive dialogue; According to the condition category, determining a target category whose target value meets the set condition in each historical conversation, the target category being the category to which the response expression located after the condition category in each historical conversation belongs; Determining a response statement of the machine response link according to each historical conversation of the target category whose target value meets the set conditions; The determining unit is specifically configured to: determine, in the category tree diagram, each target category corresponding to the condition category; and determine, in the historical conversations containing the condition category, a ratio of the number of historical conversations whose conversation results meet a first preset threshold to the number of historical conversations containing the condition category as a first conversation result; For any target category, in the historical conversations including the condition category and the target category, the ratio of the number of historical conversations whose conversation results meet the first preset threshold to the number of historical conversations including the condition category and the target category is determined as the second conversation result; determining a ratio of the first dialogue result to the second dialogue result as a target value of the target category; Among the target categories, a target category whose target value satisfies the set conditions is determined.

7. A computing device, characterized in that include: Memory for storing computer programs; A processor, configured to call a computer program stored in the memory, and execute the method according to any one of claims 1 to 5 according to the obtained program.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer-executable program, and the computer-executable program is used to enable a computer to execute the method according to any one of claims 1 to 5.

Images