Method for constructing automatic response model and automatic response method

Abstract

The present invention provides a method for constructing an automatic response model and an automatic response method. The method for constructing the automatic response model includes: extracting features from target response information to generate first identification information corresponding to the target response information, where the target response information is customer response information in the case of failed machine marketing; based on the first identification information, determining at least one target marketing script information from candidate artificial marketing script information corresponding to candidate customer response information, where the candidate customer response information is all customer response information in the case of successful artificial marketing. The method for constructing the automatic response model of the present invention can improve the conversion rate of machine marketing.

Description

Technical Field

[0001] The present invention relates to the technical field of information processing, and in particular, to a method for constructing an automatic response model and an automatic response method. Background Art

[0002] Currently, outbound robots are widely used in marketing scenarios. In related technologies, a fixed set of script information is generally set for the outbound robot to achieve a conversation with the customer. However, this method cannot automatically determine whether the relevant script is effective, and manual intervention is still required in actual execution to improve the success rate and conversion rate of marketing. This not only has low effectiveness but also consumes high labor costs and time costs. Summary of the Invention

[0003] The present invention provides a method for constructing an automatic response model and an automatic response method to solve the defect of low conversion rate of machine marketing in the prior art and improve the conversion rate of machine marketing.

[0004] The present invention provides a method for constructing an automatic response model, including:

[0005] Performing feature extraction on a target response message to generate first identification information corresponding to the target response message, where the target response message is a customer response message in the case of machine marketing failure;

[0006] Based on the first identification information, determining at least one target marketing script information from candidate artificial marketing script information corresponding to candidate customer response messages, where the candidate customer response messages are all customer response messages in the case of successful artificial marketing.

[0007] According to the method for constructing an automatic response model provided by the present invention, the determining at least one target marketing script information from candidate artificial marketing script information corresponding to candidate customer response messages based on the first identification information includes:

[0008] Screening the first identification information to generate a set of second identification information;

[0009] Removing duplicates from the target response messages corresponding to the second identification information to generate a set of first customer response messages corresponding to the second identification information, where each of the first customer response messages in the set of first customer response messages has the same meaning but different expressions;

[0010] Based on the set of first customer response messages and the previous machine marketing script information corresponding to the first customer response messages, matching at least one target marketing script information corresponding to the second identification information from the candidate artificial marketing script information.

[0011] According to a method for constructing an automatic response model provided by the present invention, the matching of obtaining at least one target marketing speech information corresponding to the second identification information from the candidate manual marketing speech information based on the set of the first customer reply information and the above machine marketing speech information corresponding to the first customer reply information includes:

[0012] Determining first similarity information between the first customer reply information and each of the candidate customer reply information;

[0013] In a case where the first similarity information exceeds a first target threshold, determining the candidate customer reply information corresponding to the first similarity information as the second customer reply information;

[0014] Determine second similarity information of the above machine marketing speech information corresponding to the first customer reply information and the above manual marketing speech information corresponding to the second customer reply information;

[0015] In the case where the second similarity information exceeds a second target threshold, the following artificial marketing speech information corresponding to the preceding artificial marketing speech information corresponding to the second similarity information is determined as the target marketing speech information.

[0016] According to the method for constructing an automatic answering model provided by the present invention, the determining of the first similarity information between the first customer reply information and the reply information of each candidate customer includes:

[0017] Extracting features from the first customer reply information to generate a first feature vector code and a first key information set;

[0018] Extracting features from the candidate customer reply information to generate a first target feature vector code and a first target key information set;

[0019] Determine first similarity information between the first customer reply information and each of the candidate customer reply information based on the first feature vector code, the first key information set, the first target feature vector code, and the first target key information set;

[0020] and / or,

[0021] The determining of second similarity information of the above machine marketing speech information corresponding to the first customer reply information and the above manual marketing speech information corresponding to the second customer reply information includes:

[0022] Extract features of the above machine marketing speech information corresponding to the first customer reply information to generate a second feature vector code and a second key information set;

[0023] Perform feature extraction on the above manual marketing speech information corresponding to the second customer reply information to generate a second target feature vector code and a second target key information set;

[0024] Based on the second feature vector code, the second key information set, the second target feature vector code and the second target key information set, second similarity information between the above machine marketing speech information and the above manual marketing speech information is determined.

[0025] According to a method for constructing an automatic answering model provided by the present invention, the step of extracting features from target reply information and generating first identification information corresponding to the target reply information includes:

[0026] Extracting features of the target reply information to generate third identification information and probability values corresponding to each of the third identification information;

[0027] When the probability value exceeds a third target threshold, the third identification information corresponding to the probability value is determined as the first identification information.

[0028] According to a method for constructing an automatic response model provided by the present invention, after determining at least one target marketing speech information from candidate manual marketing speech information corresponding to candidate customer reply information based on the first identification information, the method further includes:

[0029] Extracting features of the target marketing speech information to generate multiple type labels corresponding to the target marketing speech information;

[0030] Based on the target marketing speech structured classification system, the multiple type tags are structured to generate structured classification tags corresponding to the target marketing speech information;

[0031] Based on the structured classification tags, a target marketing strategy corresponding to the target reply information is generated.

[0032] The present invention also provides an automatic answering method, comprising:

[0033] Get the reply information of the customers waiting to be answered;

[0034] Extracting features of the reply information of the customer to be answered, and generating target identification information corresponding to the reply information of the customer to be answered;

[0035] The target identification information is input into the automatic response model generated by any of the automatic response model construction methods described above to generate target response information.

[0036] The present invention also provides a device for constructing an automatic response model, comprising:

[0037] A first generation module, configured to perform feature extraction on a target reply message to generate first identification information corresponding to the target reply message, where the target reply message is a customer reply message in the case of machine marketing failure;

[0038] A first determination module, configured to determine at least one target marketing script information from candidate artificial marketing script information corresponding to candidate customer reply messages based on the first identification information, where the candidate customer reply messages are all customer reply messages in the case of successful artificial marketing.

[0039] The present invention further provides an automatic answering device, including:

[0040] An acquisition module, configured to acquire a customer reply message to be answered;

[0041] A second generation module, configured to perform feature extraction on the customer reply message to be answered to generate target identification information corresponding to the customer reply message to be answered;

[0042] A third generation module, configured to input the target identification information into an automatic answering model generated by the method for constructing an automatic answering model according to any one of the above, to generate a target answering message.

[0043] The present invention further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where when the processor executes the computer program, the steps of the method for constructing an automatic answering model or the steps of an automatic answering method according to any one of the above are implemented.

[0044] The present invention further provides a non-transitory computer-readable storage medium, on which a computer program is stored, where when the computer program is executed by a processor, the steps of the method for constructing an automatic answering model or the steps of an automatic answering method according to any one of the above are implemented.

[0045] The present invention further provides a computer program product, including a computer program, where when the computer program is executed by a processor, the steps of the method for constructing an automatic answering model or the steps of an automatic answering method according to any one of the above are implemented.

[0046] The method for constructing an automatic response model and the automatic response method provided by the present invention extract features from target response information in the case of machine marketing failure to generate first identification information corresponding to the target response information, and based on the first identification information, determine at least one target marketing conversation information from the candidate artificial marketing conversation information corresponding to the customer response information in the case of successful artificial marketing, so as to establish an association relationship between the first identification information and the target marketing conversation information, which is helpful for quickly matching the corresponding marketing conversation information based on the customer response information in the subsequent application process, thereby improving the marketing conversion rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0048] Figure 1 is one of the flow diagrams of the method for constructing an automatic response model provided by the present invention;

[0049] Figure 2 is the second flow diagram of the method for constructing an automatic response model provided by the present invention;

[0050] Figure 3 is the third flow diagram of the method for constructing an automatic response model provided by the present invention;

[0051] Figure 4 is the fourth flow diagram of the method for constructing an automatic response model provided by the present invention;

[0052] Figure 5 is the flow diagram of the automatic response method provided by the present invention;

[0053] Figure 6 is the structural diagram of the device for constructing an automatic response model provided by the present invention;

[0054] Figure 7 is the structural diagram of the automatic response device provided by the present invention;

[0055] Figure 8 is the structural diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0056] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention with reference to the accompanying drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without creative efforts shall fall within the protection scope of the present invention.

[0057] The following will describe Figures 1 - 4 the method for constructing an automatic response model of the present invention.

[0058] It should be noted that the execution subject of the method for constructing the automatic response model of the present invention may be a device for constructing the automatic response model, or a server, or a user's terminal, such as a mobile phone or a computer, etc.

[0059] As Figure 1 shown, the method for constructing the automatic response model includes: step 110 and step 120.

[0060] Step 110: Extract features from the target response information to generate first identification information corresponding to the target response information, where the target response information is the customer response information in the case of failed machine marketing.

[0061] It should be noted that in the machine outbound call scenario, the conversation between the robot customer service and the customer can be a conversation initiated by the customer or a conversation initiated by the robot customer service.

[0062] A conversation initiated by the customer is called a voice navigation robot.

[0063] A conversation actively initiated by the robot customer service is called an outbound call robot.

[0064] During the entire process of the outbound call robot making an outbound call, it involves the machine speech uttered by the robot and the response information replied by the customer for each machine speech.

[0065] Currently, the machine outbound call scenarios are mainly divided into: notification type, return visit type, and marketing type. Among them, the notification type and the return visit type scenarios are relatively simple and the customer cooperation degree is relatively high; the marketing scenario is to recommend products to customers.

[0066] It can be understood that in the machine outbound call marketing scenario, the final marketing results include marketing failure and marketing success. For example, when the outbound call robot successfully recommends a certain product to the customer, it means marketing success; when the customer refuses to purchase the recommended product, it means marketing failure.

[0067] In this step, all call information generated during the marketing process can be converted into text information, and the target reply information is the text information corresponding to all customer reply information in the case of machine marketing failure during the historical marketing process.

[0068] It can be understood that in each case of machine marketing failure, there is at least one customer reply information.

[0069] In the actual execution process, the customer reply information in the case of machine marketing failure during the historical process can be stored in a local database or a cloud database and retrieved when needed to obtain the target reply information.

[0070] The first identification information can be a label in the form of a word or a phrase, etc., used to represent the core meaning of the target reply information.

[0071] It can be understood that different target reply information has different meanings, and their corresponding first identification information is also different.

[0072] By extracting features from multiple target reply information, multiple first identification information can be obtained. There may be the same first identification information or different first identification information among the multiple first identification information.

[0073] By classifying multiple first identification information, the first identification information under different types can be obtained.

[0074] For the first identification information in the case of machine marketing failure, it can include: multiple major category labels such as user has no demand, user's demand is uncertain, user has demand but does not meet the handling conditions, user has demand but cannot operate, and user has demand but the verification is not successful.

[0075] Each major category label can also include multiple minor category labels.

[0076] For example, taking the marketing of traffic packages as an example, under the major category label of user has no demand, it can include: minor category labels such as the user hung up directly without saying a word, the user directly refused without reason, the user expressed that the current traffic is sufficient, and the user is satisfied with the current package.

[0077] Under the major category label of user's demand is uncertain, it can include: minor category labels such as expressing temporarily not needed, expressing considering, expressing being busy, expressing sending text messages, worrying about forgetting to cancel, handling at the business hall, complaining about the high cost, and the number is not often used.

[0078] Under the major category label of user has demand but does not meet the handling conditions, it can include: minor category labels such as secondary card, bundled with other services, and non-decision maker.

[0079] Under the major category of tags where the user has a need but cannot operate, it may include sub-category tags such as not being able to press buttons, etc.

[0080] Under the major category of tags where the user has a need but has not verified and confirmed success, it may include sub-category tags such as agreeing but not pressing the confirmation button, and agreeing but pressing the wrong button, etc.

[0081] It should be noted that in the present invention, the first identification information may include, but is not limited to, multiple sub-category tags under multiple major category tags.

[0082] For example, in the case of recommending a data package plan to the user, after the call is connected, the outbound robot makes a product recommendation, and the user replies "Well, no thanks. I have enough here", then the sentence "Well, no thanks. I have enough here" is the target reply information.

[0083] Feature extraction is performed on the target reply information to obtain the key information "Data is sufficient", then "Data is sufficient" is the first identification corresponding to the target reply information.

[0084] After generating the first identification information, the first identification information can be stored in a local database or a cloud database, and can be retrieved when needed.

[0085] Step 120: Based on the first identification information, determine at least one target marketing script information from the candidate artificial marketing script information corresponding to the candidate customer reply information, where the candidate customer reply information is all the customer reply information in the case of successful artificial marketing.

[0086] In this step, the candidate customer reply information is all the reply information of the user to the artificial marketing script in the case of successful artificial marketing.

[0087] For each successful case of artificial marketing, it includes at least one piece of customer reply information.

[0088] The candidate artificial marketing script information is the context artificial marketing script information corresponding to each piece of customer reply information in the case of successful artificial marketing.

[0089] It can be understood that each piece of candidate customer reply information corresponds to at least one piece of artificial marketing script information.

[0090] It should be noted that in the actual marketing process, in the context of artificial marketing, there is a situation where after the customer refuses, the customer is persuaded through artificial marketing script information and finally successful marketing is achieved; while in the context of machine marketing, after the customer says the same reply information to refuse, the robot fails to successfully persuade the customer, resulting in marketing failure.

[0091] Table 1 shows partial call content in two different scenarios: machine marketing failure and successful manual marketing. On the left side are the customer response information and the corresponding machine marketing script information in the scenario of machine marketing failure, and on the right side are the customer response information and the corresponding manual marketing script information in the scenario of successful manual marketing.

[0092] According to Table 1, after the robot and the agent respectively finished the opening introduction, users X and Y made the same meaning responses. Then, the robot and the agent respectively sent subsequent recovery scripts based on the users' response information. User X hung up directly, resulting in the failure of machine marketing, while user Y agreed to handle it, successfully recovered, and manual marketing was successful.

[0093] For Table 1, the response information of user X is the target response information, and the label extracted based on the response information of user X is the first identification information corresponding to the response information of user X; the response information of user Y is the candidate customer response information, and the script information of the agent is the candidate manual marketing script information corresponding to the response information of user Y, and the first identification information corresponding to the response information of user X has the same or similar actual meaning as the response information of user Y.

[0094] Table 1

[0095]

[0096] In this step, by processing the first identification information and its corresponding target response information, the target marketing script information corresponding to the first identification information is matched, so that the association relationship between the first identification information and the target marketing script information can be established.

[0097] According to the method for constructing an automatic response model provided by the embodiment of the present invention, by extracting features from the target response information in the case of machine marketing failure, the first identification information corresponding to the target response information is generated, and based on the first identification information, at least one target marketing script information is determined from the candidate manual marketing script information corresponding to the customer response information in the case of successful manual marketing, so that the association relationship between the first identification information and the target marketing script information can be established, which helps to quickly match the corresponding marketing script information based on the customer response information in the subsequent application process, thereby improving the marketing conversion rate; in addition, it can also realize the closed-loop optimization from human-machine interaction analysis to human-human dialogue analysis.

[0098] In some embodiments, step 110 includes:

[0099] Extract features from the target response information to generate the third identification information and the probability value corresponding to each third identification information;

[0100] When the probability value exceeds the third target threshold, the third identification information corresponding to the probability value is determined as the first identification information.

[0101] In this embodiment, the third identification information may be a tag in the form of a word or a phrase, etc., and is used to characterize the possible meaning included in the target reply information.

[0102] It can be understood that for the same target reply information, one or more third identification information can be extracted.

[0103] The probability value corresponding to the third identification information is used to characterize the similarity degree between the third identification information and the core meaning included in the target reply information.

[0104] The higher the probability value, the closer the third identification information is to the core meaning included in the target reply information.

[0105] The third target threshold is an evaluation criterion for determining whether the third identification information can be used to characterize the target reply information. The third target threshold can be user-defined, such as set to 80% or 70%, etc.

[0106] When the probability value corresponding to the third identification information exceeds the third target threshold, it can be approximately considered that the third identification information can be used to characterize the meaning to be expressed by the target reply information, and then the third identification information is determined as the first identification information.

[0107] In the actual execution process, the BERT model in natural language processing can be used to add a classification layer Softmax function to predict the first identification information corresponding to the target reply information.

[0108] Among them, BERT (Bidirectional Encoder Representation from Transformers) is a pre-trained language representation model for generating deep bidirectional language representations.

[0109] The Softmax function, also known as the normalized exponential function, is used to display the results of multi-classification in the form of probabilities.

[0110] The specific network structure is as Figure 2 shown. Among them, Wi represents the i-th character corresponding to the text of the target reply information, [CLS] and [SEP] are special identifiers added at the beginning and end of the text of the target reply information, Ei is the initial encoding corresponding to the i-th character in the text of the target reply information, and Hi is the encoding of the i-th character after passing through the BERT model.

[0111] The core idea of the entire network structure is: first, use the BERT model to obtain the feature representation corresponding to the target reply information, and then pass the feature representation of the user text CLS position corresponding to each sentence of the target reply information through the fully connected layer and the Softmax function to obtain the probability values corresponding to all label categories, that is, to obtain the third identification information and the probability values corresponding to each third identification information.

[0112] By comparing the probability value with the third target threshold, if the probability value of the label category is not lower than the preset third target threshold, it means that the probability value corresponding to the label category is larger, and the corresponding label is output, which is the first identification information.

[0113] In some embodiments, when the probability value corresponding to the third identification information is lower than the third target threshold, an unknown label is output.

[0114] According to the method for constructing an automatic answering model provided by an embodiment of the present invention, by performing feature extraction on target reply information and screening based on probability values to obtain first identification information corresponding to each target reply information, the extraction accuracy is high and the first identification information finally obtained has high accuracy.

[0115] In some embodiments, step 120 includes:

[0116] Screening the first identification information to generate a set of second identification information;

[0117] De-duplication of the target reply information corresponding to the second identification information is performed to generate a set of first customer reply information corresponding to the second identification information, wherein each first customer reply information in the set of first customer reply information has the same meaning but different wordings;

[0118] Based on the set of first customer reply information and the above machine marketing speech information corresponding to the first customer reply information, at least one target marketing speech information corresponding to the second identification information is matched from the candidate manual marketing speech information.

[0119] In this embodiment, the second identification information is a tag in the first identification information that has a higher user proportion or is more closely associated with the user.

[0120] The amount of the second identification information should not exceed the amount of the first identification information.

[0121] During the actual implementation process, the first identification information can be screened based on the user proportion corresponding to each first identification information or the association between each first identification information and the user, and the first identification information with a higher user proportion and / or a greater impact on the user can be selected as the second identification information to generate a set including multiple second identification information.

[0122] It can be understood that each type of second identification information corresponds to at least one target reply message, and the target reply messages corresponding to the same type of second identification information have the same meaning but may be expressed differently.

[0123] Table 2 shows two types of second identification information and target reply information corresponding to each type of second identification information.

[0124] Table 2

[0125]

[0126] During the actual implementation process, the multiple target reply information corresponding to each type of second identification information is screened respectively, basically similar statements are deleted, and user statements that express the same meaning but have significant differences in statement are retained to be determined as first customer reply information. The multiple first customer reply information corresponding to each type of second identification information constitute a set of customer reply information corresponding to this type of second identification information.

[0127] After obtaining the first customer reply information, the above machine marketing speech information corresponding to the first customer reply information is matched based on the first customer reply information.

[0128] It can be understood that each first customer reply message corresponds to at least one sentence of the above machine marketing speech information.

[0129] After obtaining the first customer reply information and its corresponding above machine marketing speech information, based on the similarity between the first customer reply information and the candidate customer reply information, as well as the similarity between the above machine marketing speech information corresponding to the first customer reply information and the above manual marketing speech information corresponding to the candidate customer reply information, the candidate manual marketing speech information that is highly correlated with the first customer reply information is matched from the candidate manual marketing speech information as the target marketing speech information corresponding to the second identification information.

[0130] It should be noted that the target marketing speech information corresponding to the second identification information may be one or more.

[0131] In this embodiment, by pre-processing the first identification information and the target reply information, the amount of data in the subsequent data processing process can be effectively reduced, thereby improving the data processing rate.

[0132] The following is a specific example of how to determine the similarity in this application.

[0133] In some embodiments, matching at least one target marketing script information corresponding to the second identification information from the candidate manual marketing script information based on the set of first customer response information and the above-mentioned machine marketing script information corresponding to the first customer response information in step 120 includes:

[0134] Determine the first similarity information between the first customer response information and each candidate customer response information;

[0135] When the first similarity information exceeds the first target threshold, determine the candidate customer response information corresponding to the first similarity information as the second customer response information;

[0136] Determine the second similarity information between the above-mentioned machine marketing script information corresponding to the first customer response information and the above-mentioned manual marketing script information corresponding to the second customer response information;

[0137] When the second similarity information exceeds the second target threshold, determine the following manual marketing script information corresponding to the above-mentioned manual marketing script information corresponding to the second similarity information as the target marketing script information.

[0138] In this embodiment, the first similarity information is used to characterize the similarity degree between the first customer response information and the candidate customer response information.

[0139] The first target threshold can be user-defined, and this application does not make any limitations.

[0140] When the second customer response information is successfully manually marketed, it is a customer response information with a relatively high similarity degree to the first customer response information.

[0141] After obtaining the first similarity information, compare the first similarity information with the first target threshold. When the first similarity information exceeds the first target threshold, it is approximately considered that the similarity degree between the first customer response information corresponding to the first similarity information and the candidate customer response information is relatively high, and then determine the candidate customer response information corresponding to the first similarity information as the second customer response information.

[0142] It can be understood that multiple first customer response information can correspond to the same type of second identification information, and each first customer response information can correspond to multiple second customer response information.

[0143] Among them, the first customer response information and the second customer response information have the same or similar statements; in the same application scenario, their expressed meanings may be the same; but in different application scenarios, their actual expressed meanings may be different.

[0144] As shown in Table 3, there is a situation where the statements of customer response information are the same, but the actual expressed meanings are different.

[0145] Table 3

[0146]

[0147] In an embodiment of the present application, after obtaining the second customer reply information, the second customer reply information may also be screened, which may be specifically performed as follows:

[0148] Determine second similarity information between the above machine marketing speech information corresponding to the first customer reply information and the above manual marketing speech information corresponding to the second customer reply information;

[0149] When the second similarity information exceeds the second target threshold, the following artificial marketing speech information corresponding to the preceding artificial marketing speech information corresponding to the second similarity information is determined as the target marketing speech information.

[0150] Among them, the second similarity information is used to characterize the similarity between the above machine marketing speech information corresponding to the first customer reply information and the above manual marketing speech information corresponding to the second customer reply information.

[0151] The target marketing speech information is the following manual marketing speech information corresponding to the second customer's reply information when the manual marketing is successful.

[0152] The second target threshold can be customized by the user and is not limited in this application.

[0153] After obtaining the second similarity information, compare the second similarity information with the second target threshold. When the second similarity information exceeds the second target threshold, it is approximately considered that the degree of similarity between the above machine marketing speech corresponding to the first customer reply information corresponding to the second similarity information and the above manual marketing speech corresponding to the second customer reply information is relatively high, that is, the application scenarios between the first customer reply information and the second customer reply information are approximately the same, then the below manual marketing speech information corresponding to the above manual marketing speech information corresponding to the second similarity information is determined as the target marketing speech information.

[0154] Finally, based on the target reply information corresponding to the second identification information and its previous machine marketing speech information, similar user statements in the case of successful manual marketing and its following manual marketing speech information are mined and determined as the corresponding target marketing speech information corresponding to the second identification information.

[0155] The steps of generating the first similarity information and the second similarity information are described below through a specific embodiment.

[0156] In some embodiments, determining first similarity information between the first customer reply information and each candidate customer reply information includes:

[0157] Extract features from the first customer response information to generate a first feature vector encoding and a first set of key information;

[0158] Extract features from the candidate customer response information to generate a first target feature vector encoding and a first target set of key information;

[0159] Based on the first feature vector encoding, the first set of key information, the first target feature vector encoding, and the first target set of key information, determine the first similarity information between the first customer response information and each candidate customer response information.

[0160] In this embodiment, the first feature vector encoding corresponding to the first customer response information is the feature vector encoding at the text character level of the first customer response information;

[0161] The first set of key information corresponding to the first customer response information is the words expressing the key information of sentences or semantic fragments in the text of the first customer response information.

[0162] In the actual execution process, the BERT model can be used to extract the feature vector at the text character level of the first customer response information to generate the first feature vector encoding.

[0163] Use the BERT+ATT model to extract key tokens from the first customer response information to generate the first set of key information. Among them, the categories of Tokens are generally divided into skill categories, business categories, modification categories, and sentence pattern categories.

[0164] The skill category is the operation words related to the business, such as query, handle, and change, etc.; the business category is the specific business objects in the field, such as traffic, phone bill, and data package, etc.; the modification category is the attributes included in the business product, such as specific amount, specific traffic, exceeding, last month, next month, main card, sub-card, and sharing, etc.; the sentence pattern category is the category expressing the sentence, such as statement and question, etc.

[0165] For example, for the first customer response information 1: Am I supposed to handle a data package?, the corresponding extracted Tokens are: handle#data package#question; for the first customer response information 2: Check how much traffic I have left, the corresponding extracted Tokens are: query#traffic#question.

[0166] As Figure 3 shown, taking the first customer response information as the user statement X as an example, use the BERT model to extract the feature vector at the text character level of the user statement X to generate the first feature vector encoding Ex; use the BERT+ATT model to extract key tokens from the user statement X to generate the first set of key information Tx.

[0167] According to the same method, feature extraction can be performed on the candidate customer response information to generate the first target feature vector encoding and the first target key information set corresponding to the candidate customer response information, which will not be elaborated here.

[0168] Among them, the first target feature vector encoding is the feature vector encoding at the text character level of the candidate customer response information;

[0169] The first target key information set is the words expressing the sentence key information or semantic fragments in the text of the candidate customer response information.

[0170] After generating the first feature vector encoding, the first key information set, the first target feature vector encoding, and the first target key information set, based on the first feature vector encoding and the first target feature vector encoding, the first sub-similarity information between the first customer response information and each candidate customer response information at the sentence text vector level can be generated; based on the first key information set and the first target key information set, the second sub-similarity information between the first customer response information and each candidate customer response information at the Token level can be generated.

[0171] During the actual execution process, the formula can be used:

[0172]

[0173] to generate the first sub-similarity information, where Ex1 is the first feature vector encoding of the first customer response information, Ex2 is the first target feature vector encoding of the candidate customer response information, and Sim_vector(Ex1, Ex2) is the first sub-similarity information.

[0174] It can be understood that the smaller the value of the first sub-similarity information, the smaller the similarity.

[0175] Through the formula:

[0176]

[0177] to generate the second sub-similarity information, where Tx1 is the first key information set of the first customer response information, Tx2 is the first target key information set of the candidate customer response information, β is the harmonic factor, β ranges from 0 to 1, the Levenshtein distance refers to the minimum number of edit operations required to convert one string to another between two strings, where len is the word length corresponding to the key token segment, max is to take the maximum value; Sim_token(Tx1, Tx2) is the second sub-similarity information.

[0178] Similarly, the smaller the value of the second sub-similarity information, the smaller the similarity.

[0179] The first similarity information may be generated by performing a weighted fusion algorithm calculation on the first sub-similarity information and the second sub-similarity information.

[0180] In the actual implementation process, the formula can be used:

[0181]

[0182] Generate the first similarity information, where X1 is the first customer reply information, X2 is the candidate customer reply information, S(X1, X2) is the weighted fusion score of the vector metric and the key token metric, that is, the first similarity information, and α is the reconciliation factor, and α takes a value between 0 and 1.

[0183] It should be noted that, when performing weighted fusion algorithm calculation on the first sub-similarity information and the second sub-similarity information, the weight α of the first sub-similarity and the second sub-similarity can be user-defined, or adopt a default value, or can be continuously optimized in subsequent calculation processes.

[0184] For example, during the initial calculation, the weight value can be set to 1:1. In the subsequent training process, based on the importance of the sentence text vector level and the token level, or based on the output results, the weight value used in the previous training is continuously optimized and adjusted to ultimately determine the optimal weight value.

[0185] In some embodiments, determining second similarity information between the above machine marketing speech information corresponding to the first customer reply information and the above manual marketing speech information corresponding to the second customer reply information includes:

[0186] Extract features of the above machine marketing speech information corresponding to the first customer reply information to generate a second feature vector code and a second key information set;

[0187] Perform feature extraction on the above manual marketing speech information corresponding to the second customer reply information to generate a second target feature vector code and a second target key information set;

[0188] Based on the second feature vector code, the second key information set, the second target feature vector code and the second target key information set, second similarity information between the above machine marketing speech information and the above manual marketing speech information is determined.

[0189] In this embodiment, feature extraction is performed on the above machine marketing speech information corresponding to the first customer's reply information to generate a second feature vector code and a second key information set corresponding to the above machine marketing speech information; feature extraction is performed on the above manual marketing speech information corresponding to the second customer's reply information to generate a second target feature vector code and a second target key information set corresponding to the above manual marketing speech information.

[0190] The second feature vector code is a feature vector code at the text word level of the above machine marketing speech information corresponding to the first customer reply information;

[0191] The second key information set is words that express key information of sentences or semantic fragments in the text of the above machine marketing speech information corresponding to the first customer reply information.

[0192] The second target feature vector encoding is a feature vector encoding at the text word level of the above manual marketing speech information corresponding to the second customer reply information;

[0193] The second target key information set dimension expresses the key information of sentences or words of semantic fragments in the text of the above manual marketing speech information corresponding to the second customer reply information.

[0194] The specific feature extraction method is the same as that in the above embodiment, and can be implemented using the BERT model and the BERT+ATT model, which will not be described in detail here.

[0195] After generating the second feature vector code, the second key information set, the second target feature vector code and the second target key information set, based on the second feature vector code and the second target feature vector code, the third sub-similarity information at the sentence text vector level between the above machine marketing speech information corresponding to the first customer reply information and the above manual marketing speech information corresponding to each second customer reply information can be generated; based on the second key information set and the second target key information set, the fourth sub-similarity information at the Token level between the above machine marketing speech information corresponding to the first customer reply information and the above manual marketing speech information corresponding to each second customer reply information can be generated.

[0196] The calculation method of the third sub-similarity and the fourth sub-similarity is the same as that in the above embodiment, and will not be described in detail here.

[0197] The second similarity information may be generated by performing a weighted fusion algorithm calculation on the third sub-similarity information and the fourth sub-similarity information.

[0198] The method for calculating the second similarity information is the same as that in the above embodiment and will not be described in detail here.

[0199] It should be noted that, when performing weighted fusion algorithm calculation on the third sub-similarity information and the fourth sub-similarity information, the weights of the third sub-similarity and the fourth sub-similarity may be user-defined, or adopt default values, or may be continuously optimized in subsequent calculation processes.

[0200] According to the method for constructing an automatic answering model provided by an embodiment of the present invention, candidate customer reply information similar to the target reply information is determined through a first similarity, and the previous manual marketing speech information similar to the previous machine marketing speech information corresponding to the target reply information is determined from the previous manual marketing speech information corresponding to these candidate customer reply information through a second similarity, thereby ensuring that the previous marketing scenarios and strategies corresponding to the obtained similar user statements are also basically consistent; on this basis, an association relationship based on the second identification information is established between the target reply information whose customer reply information is similar to the previous marketing speech information and the following manual marketing speech information corresponding to the candidate reply information, thereby effectively improving the precision and accuracy of the results.

[0201] like Figure 4 As shown, according to some embodiments of the present invention, after step 120, the method further includes:

[0202] Extract features of target marketing speech information and generate multiple type labels corresponding to the target marketing speech information;

[0203] Based on the target marketing speech structured classification system, multiple type tags are structured and processed to generate structured classification tags corresponding to the target marketing speech information;

[0204] Based on structured classification labels, target marketing strategies corresponding to target response information are generated.

[0205] In this embodiment, the type tag may be a keyword or phrase in the target marketing speech information, which is used to characterize the core meaning of the target marketing speech information.

[0206] Table 4 shows the correspondence table of several target reply information and their corresponding type labels.

[0207] As shown in Table 4, for the target marketing speech information 1, multiple corresponding type tags can be generated, such as expensive traffic overflow, cost-effective marketing activities, introduction to traffic usage scenarios, and first experience, etc.

[0208] Then, based on the structured classification system of the target speech, multiple type labels are structured to generate the order and logical relationship between the types of labels, thereby generating structured classification labels.

[0209] In the actual implementation process, a predefined structured classification system for speech can be adopted. Based on the BERT+ATT model in natural language processing and using the sliding window idea, each sentence in the target marketing speech information corresponding to each second identification information can be assigned to a category, thereby obtaining multiple type labels corresponding to each target marketing speech information after structuring.

[0210] After generating the structured classification tags, based on the structured classification tags, the target marketing strategy corresponding to the target reply information can be generated.

[0211] It can be understood that each target marketing speech information corresponds to multiple structured classification tags.

[0212] On this basis, the multiple structured classification tags corresponding to each target marketing speech information are summarized and statistically analyzed, and then the specific strategies corresponding to different marketing speeches can be obtained.

[0213] Table 4

[0214]

[0215] For example, Strategy 1: Compare prices # Experience first; Strategy 2: High traffic cost # Cost-effective marketing activities. Each strategy also has diverse speech content.

[0216] In addition, for each target marketing speech information obtained by mining, combined with its context multi-round interaction record information, the marketing path is summarized and analyzed. Not only can the excellent marketing strategies and speech content corresponding to the second identification information that has not been successfully marketed currently be obtained, but also the corresponding combined marketing strategies and speech content can be obtained.

[0217] In this application, based on the historical interaction data of users who have not been successfully marketed by the machine, excellent marketing speeches and marketing strategies of gold medal agents can be mined. In the subsequent process of applying them to the optimization of the process and speech of the machine outbound marketing scenario, not only can the user interaction experience be improved, but also the machine marketing conversion rate can be increased, and the revenue can be increased.

[0218] According to the method for constructing an automatic response model provided by the embodiment of the present invention, by extracting features from the target marketing speech information to generate structured classification tags, and generating target marketing strategies based on the structured classification tags, on the basis of mining excellent marketing speeches, excellent marketing strategies can be further mined, so that not only can the excellent marketing strategies and excellent marketing speech information corresponding to the first identification information of the user reply information that has not been successfully marketed currently be obtained, but also the corresponding combined marketing strategies and speech content can be obtained, thereby helping to discover new strategies that are difficult to think of by artificial experience and ideas, effectively solving problems such as difficult discovery of marketing strategies, lack of pertinence in speech optimization, and inflexible marketing.

[0219] Next, the device for constructing an automatic response model provided by the present invention will be described. The device for constructing an automatic response model described below can be correspondingly referred to the method for constructing an automatic response model described above.

[0220] Such as Figure 6As shown, the apparatus for constructing an automatic response model includes: a first generation module 610 and a first determination module 620.

[0221] The first generation module 610 is configured to extract features from the target reply information to generate first identification information corresponding to the target reply information, where the target reply information is customer reply information in the case of failed machine marketing.

[0222] The first determination module 620 is configured to determine at least one target marketing script information from the candidate artificial marketing script information corresponding to the candidate customer reply information based on the first identification information, where the candidate customer reply information is all customer reply information in the case of successful artificial marketing.

[0223] According to the apparatus for constructing an automatic response model provided by an embodiment of the present invention, by extracting features from the target reply information in the case of failed machine marketing to generate first identification information corresponding to the target reply information, and based on the first identification information, determining at least one target marketing script information from the candidate artificial marketing script information corresponding to the customer reply information in the case of successful artificial marketing, an association relationship between the first identification information and the target marketing script information can be established, which helps to quickly match the corresponding marketing script information based on the customer reply information in the subsequent application process, thereby improving the marketing conversion rate.

[0224] In some embodiments, the first determination module 620 is further configured to:

[0225] Screen the first identification information to generate a set of second identification information;

[0226] Deduplicate the target reply information corresponding to the second identification information to generate a set of first customer reply information corresponding to the second identification information, where each first customer reply information in the set of first customer reply information has the same meaning but different expressions;

[0227] Based on the set of first customer reply information and the above-mentioned machine marketing script information corresponding to the first customer reply information, match at least one target marketing script information corresponding to the second identification information from the candidate artificial marketing script information.

[0228] In some embodiments, the first determination module 620 is further configured to:

[0229] Determine first similarity information between the first customer reply information and each candidate customer reply information;

[0230] In the case where the first similarity information exceeds the first target threshold, determine the candidate customer reply information corresponding to the first similarity information as the second customer reply information;

[0231] Determine second similarity information between the above machine marketing speech information corresponding to the first customer reply information and the above manual marketing speech information corresponding to the second customer reply information;

[0232] When the second similarity information exceeds the second target threshold, the following artificial marketing speech information corresponding to the preceding artificial marketing speech information corresponding to the second similarity information is determined as the target marketing speech information.

[0233] In some embodiments, the first determining module 620 is further configured to:

[0234] Extracting features of the first customer reply information to generate a first feature vector code and a first key information set;

[0235] Extracting features from the candidate customer reply information to generate a first target feature vector code and a first target key information set;

[0236] Based on the first feature vector code, the first key information set, the first target feature vector code and the first target key information set, first similarity information between the first customer reply information and each candidate customer reply information is determined.

[0237] In some embodiments, the first determining module 620 is further configured to:

[0238] Extract features of the above machine marketing speech information corresponding to the first customer reply information to generate a second feature vector code and a second key information set;

[0239] Perform feature extraction on the above manual marketing speech information corresponding to the second customer reply information to generate a second target feature vector code and a second target key information set;

[0240] Based on the second feature vector code, the second key information set, the second target feature vector code and the second target key information set, second similarity information between the above machine marketing speech information and the above manual marketing speech information is determined.

[0241] In some embodiments, the first generating module 610 is further used to

[0242] Extracting features of the target reply information to generate third identification information and probability values corresponding to each third identification information;

[0243] When the probability value exceeds the third target threshold, the third identification information corresponding to the probability value is determined as the first identification information.

[0244] In some embodiments, the apparatus further comprises:

[0245] The fourth generation module is used to extract features from the target marketing script information after determining at least one target marketing script information from the candidate artificial marketing script information corresponding to the candidate customer response information based on the first identification information, and generate multiple type tags corresponding to the target marketing script information;

[0246] The fifth generation module is used to perform structured processing on the multiple type tags based on the target script structured classification system to generate structured classification tags corresponding to the target marketing script information;

[0247] The sixth generation module is used to generate a target marketing strategy corresponding to the target response information based on the structured classification tags.

[0248] The automatic response method provided by the present invention will be described below. The automatic response method described below can be referred to in correspondence with the construction method of the automatic response model described above.

[0249] It should be noted that the execution subject of this automatic response method can be an automatic response device, or a server, or also a user's terminal, such as a mobile phone or a computer, etc.

[0250] As Figure 5 shown, this automatic response method includes: Step 510, Step 520, and Step 530.

[0251] Step 510: Obtain the customer response information to be replied;

[0252] In this step, the customer response information to be replied is the customer response information collected in real time during the machine marketing process.

[0253] Step 520: Extract features from the customer response information to be replied to generate target identification information corresponding to the customer response information to be replied;

[0254] In this step, the target identification information can be a label in the form of a word or a phrase, etc., used to represent the core meaning of the customer response information to be replied

[0255] After obtaining the real-time customer response information to be replied, features can be extracted from the customer response information to be replied first to generate multiple candidate identification information and the probability values corresponding to each candidate identification information; then based on the probability values, the candidate identification information with a higher probability value is determined as the target identification information.

[0256] The specific implementation method is the same as that of the above embodiment. For example, the BERT model in natural language processing can be used to add a classification layer Softmax function to predict the target identification information corresponding to the customer response information to be replied, which will not be elaborated here.

[0257] Step 530: Input the target identification information into the automatic response model generated by the method for constructing the automatic response model as described above to generate target response information.

[0258] In this step, the automatic response model is the model generated by the method for constructing the automatic response model, and is used to generate the target response information corresponding to the target identification information based on the target identification information.

[0259] Among them, the target response information is the subsequent artificial marketing script information corresponding to the customer response information similar to the target response information in the case of successful artificial marketing.

[0260] Input the target identification information generated in step 520 into the automatic response model, and the target response information corresponding to the target identification information can be obtained by matching.

[0261] According to the automatic response method provided by the embodiments of the present invention, the target identification information is generated by extracting the characteristics of the customer response information to be answered in the case of machine marketing, and then the target identification information is matched based on the automatic response model, and the subsequent artificial marketing script information in the case of successful artificial marketing corresponding to the target identification information is obtained by matching, so that the best marketing script can be generated based on the real-time response information of the customer, the matching rate is fast and the accuracy of the matching result is high, which helps to improve the conversion rate of machine marketing.

[0262] The automatic response device provided by the present invention will be described below. The automatic response device described below can be mutually referred to the automatic response method described above.

[0263] As shown in FIG. 7, the automatic response device includes: an acquisition module 710, a second generation module 720, and a third generation module 730.

[0264] The acquisition module 710 is configured to acquire the customer response information to be answered;

[0265] The second generation module 720 is configured to extract the characteristics of the customer response information to be answered to generate the target identification information corresponding to the customer response information to be answered;

[0266] The third generation module 730 is configured to input the target identification information into the automatic response model generated by the method for constructing the automatic response model as described in any one of the above to generate the target response information.

[0267] According to the automatic answering device provided by an embodiment of the present invention, by extracting features from the reply information of a customer to be answered in the case of machine marketing to generate target identification information, and then matching the target identification information based on an automatic answering model, the subsequent artificial marketing speech information in the case of successful artificial marketing corresponding to the target identification information is obtained. Thus, the best marketing speech can be generated based on the real-time reply information of the customer, with a fast matching rate and high accuracy of the matching result, which helps to improve the conversion rate of machine marketing.

[0268] Figure 8 An example of a schematic physical structure diagram of an electronic device is as Figure 8 shown. The electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840. Among them, the processor 810, the communication interface 820, and the memory 830 complete mutual communication through the communication bus 840. The processor 810 may call logic instructions in the memory 830 to execute a method for constructing an automatic answering model. The method includes: extracting features from target reply information to generate first identification information corresponding to the target reply information, where the target reply information is customer reply information in the case of failed machine marketing; determining at least one target marketing speech information from candidate artificial marketing speech information corresponding to candidate customer reply information based on the first identification information, where the candidate customer reply information is all customer reply information in the case of successful artificial marketing; or, executing an automatic answering method, the method includes: obtaining customer reply information to be answered; extracting features from the customer reply information to be answered to generate target identification information corresponding to the customer reply information to be answered; inputting the target identification information into an automatic answering model generated by the method for constructing an automatic answering model as described above to generate target answering information.

[0269] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0270] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the method for constructing an automatic response model provided by the above-mentioned methods, which includes: performing feature extraction on the target reply information, and generating first identification information corresponding to the target reply information, wherein the target reply information is the customer reply information when machine marketing fails; based on the first identification information, determining at least one target marketing speech information from the candidate manual marketing speech information corresponding to the candidate customer reply information, wherein the candidate customer reply information is all the customer reply information when manual marketing succeeds; or, executing the automatic response method, which includes: obtaining customer reply information to be answered; performing feature extraction on the customer reply information to be answered, and generating target identification information corresponding to the customer reply information to be answered; inputting the target identification information into the automatic response model generated by the method for constructing the automatic response model as described above, and generating the target response information.

[0271] In another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is configured to execute the method for constructing the automatic response model provided above. The method includes: extracting features of the target response information to generate first identification information corresponding to the target response information, where the target response information is customer response information in the case of failed machine marketing; based on the first identification information, determining at least one target marketing script information from the candidate artificial marketing script information corresponding to the candidate customer response information, where the candidate customer response information is all customer response information in the case of successful artificial marketing; or, executing an automatic response method, which includes: obtaining the customer response information to be replied; extracting features of the customer response information to be replied to generate target identification information corresponding to the customer response information to be replied; inputting the target identification information into the automatic response model generated by the method for constructing the automatic response model as described above to generate a target response information.

[0272] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative effort.

[0273] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solutions, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0274] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing an automatic response model, characterized in that include: Extracting features of target reply information to generate first identification information corresponding to the target reply information, wherein the target reply information is customer reply information when machine marketing fails; the first identification information includes multiple sub-category labels under multiple major category labels; Based on the first identification information, at least one target marketing speech information is determined from the candidate manual marketing speech information corresponding to the candidate customer reply information, wherein the candidate customer reply information is all customer reply information when the manual marketing is successful; The determining, based on the first identification information, at least one target marketing speech information from the candidate manual marketing speech information corresponding to the candidate customer reply information includes: Screening the first identification information to generate a set of second identification information; Deduplication of the target reply information corresponding to the second identification information is performed to generate a set of first customer reply information corresponding to the second identification information, wherein each of the first customer reply information in the set of the first customer reply information has the same meaning but different wordings; Based on the set of the first customer reply information and the above machine marketing speech information corresponding to the first customer reply information, at least one target marketing speech information corresponding to the second identification information is matched from the candidate manual marketing speech information.

2. The method for constructing an automatic response model according to claim 1, wherein The matching of obtaining at least one target marketing speech information corresponding to the second identification information from the candidate manual marketing speech information based on the set of the first customer reply information and the above machine marketing speech information corresponding to the first customer reply information includes: Determining first similarity information between the first customer reply information and each of the candidate customer reply information; In a case where the first similarity information exceeds a first target threshold, determining the candidate customer reply information corresponding to the first similarity information as the second customer reply information; Determine second similarity information of the above machine marketing speech information corresponding to the first customer reply information and the above manual marketing speech information corresponding to the second customer reply information; In the case where the second similarity information exceeds a second target threshold, the following artificial marketing speech information corresponding to the preceding artificial marketing speech information corresponding to the second similarity information is determined as the target marketing speech information.

3. The method for constructing an automatic response model according to claim 2, characterized in that: The determining of first similarity information between the first customer reply information and each of the candidate customer reply information includes: Extracting features from the first customer reply information to generate a first feature vector code and a first key information set; Extracting features from the candidate customer reply information to generate a first target feature vector code and a first target key information set; Determine first similarity information between the first customer reply information and each of the candidate customer reply information based on the first feature vector code, the first key information set, the first target feature vector code, and the first target key information set; and / or, The second similarity information for determining the above-mentioned machine marketing script information corresponding to the first customer reply information and the above-mentioned human marketing script information corresponding to the second customer reply information includes: Performing feature extraction on the above-mentioned machine marketing script information corresponding to the first customer reply information to generate a second feature vector encoding and a second key information set; Performing feature extraction on the above-mentioned human marketing script information corresponding to the second customer reply information to generate a second target feature vector encoding and a second target key information set; Based on the second feature vector encoding, the second key information set, the second target feature vector encoding, and the second target key information set, determining the second similarity information between the above-mentioned machine marketing script information and the above-mentioned human marketing script information.

4. The method for constructing an automatic response model according to any one of claims 1-3, characterized in that, The performing feature extraction on the target reply information to generate the first identification information corresponding to the target reply information includes: Performing feature extraction on the target reply information to generate a third identification information and probability values corresponding to each of the third identification information; In the case where the probability value exceeds a third target threshold, determining the third identification information corresponding to the probability value as the first identification information.

5. The method for constructing an automatic response model according to any one of claims 1-3, characterized in that After determining at least one target marketing script information from the candidate human marketing script information corresponding to the candidate customer reply information based on the first identification information, the method further includes: Performing feature extraction on the target marketing script information to generate multiple type labels corresponding to the target marketing script information; Based on a target script structured classification system, performing structured processing on the multiple type labels to generate a structured classification label corresponding to the target marketing script information; Based on the structured classification label, generating a target marketing strategy corresponding to the target reply information.

6. An automatic answering method, characterized in that, Including: Obtaining a customer reply information to be answered; Performing feature extraction on the customer reply information to be answered to generate a target identification information corresponding to the customer reply information to be answered; Inputting the target identification information into an automatic reply model generated by the method for constructing an automatic reply model according to any one of claims 1-4 to generate a target reply information.

7. An apparatus for constructing an automatic response model, characterized in that, Including: A first generation module, configured to perform feature extraction on a target reply information to generate a first identification information corresponding to the target reply information, where the target reply information is a customer reply information in the case of machine marketing failure; the first identification information includes multiple subclass labels under multiple major class labels; A first determination module, configured to determine at least one target marketing script information from the candidate human marketing script information corresponding to the candidate customer reply information based on the first identification information, where the candidate customer reply information is all customer reply information in the case of successful human marketing; The determining at least one target marketing script information from the candidate human marketing script information corresponding to the candidate customer reply information based on the first identification information includes: Screening the first identification information to generate a set of second identification information; Deduplication of the target reply information corresponding to the second identification information is performed to generate a set of first customer reply information corresponding to the second identification information, wherein each of the first customer reply information in the set of the first customer reply information has the same meaning but different wordings; Based on the set of the first customer reply information and the above machine marketing speech information corresponding to the first customer reply information, at least one target marketing speech information corresponding to the second identification information is matched from the candidate manual marketing speech information.

8. An automatic answering device, characterized in that, include: The acquisition module is used to obtain the reply information of the customer to be answered; The second generating module is used to extract features of the reply information of the customer to be answered, and generate target identification information corresponding to the reply information of the customer to be answered; The third generating module is used to input the target identification information into the automatic response model generated by the automatic response model construction method according to any one of claims 1 to 4 to generate target response information.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, the steps of the method for constructing the automatic response model according to any one of claims 1 to 5 or the steps of the automatic response method according to claim 6 are implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, the computer program implements the steps of the method for constructing an automatic response model according to any one of claims 1 to 5 or the steps of the automatic response method according to claim 6.

11. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, the computer program implements the steps of the method for constructing an automatic response model according to any one of claims 1 to 5 or the steps of the automatic response method according to claim 6.

Images