Problem pushing method and device applied to target scene, equipment and readable medium

Abstract

The embodiment of the invention discloses a problem pushing method and device applied to a target scene, equipment and a readable medium. A specific embodiment of the method comprises the following steps: in response to receiving user dialogue information, performing text preprocessing on the user dialogue information to obtain processed dialogue information; performing entity intention recognition on the processed dialogue information to generate dialogue entity intention information; dialogue strategy information is generated according to the dialogue entity information and the dialogue intention information; according to the dialogue entity information, the dialogue intention information and the dialogue strategy information, dialogue data is selected from a pre-constructed dialogue database to serve as target dialogue data; generating a recommended question text information sequence according to the determined target dialogue data; and pushing the recommendation question to the user side according to the generated recommendation question text information sequence. According to the embodiment, the hit rate of the pushed questions can be improved, so that the customer service dialogue experience of the user is improved, and the credibility of the pushed questions is improved.

Description

TECHNICAL FIELD

[0001] Embodiments of the present disclosure relate to the technical field of computer technology, and particularly relate to a question pushing method and device applied to a target scene, equipment and a readable medium. BACKGROUND

[0002] Question pushing refers to a technology of recommending a current inquiry question or a related question derived from a current dialogue to a user through user background information or related knowledge base content, and is usually used in a customer service dialogue scene (such as a user consultation process or an article attribute). At present, when performing question recommendation, a commonly used manner is as follows: a traditional dialogue database is constructed based on a large number of question and answer pairs, and when a user inquires a question, a question and answer pair with a higher relevance is matched from the dialogue database as a pushing question or a pushing question is generated in real time by using a large language model.

[0003] However, when the above manner is used to perform question pushing, the following technical problems often exist:

[0004] When a traditional dialogue database is used to perform question matching, the matched pushing question only has a higher semantic similarity with the user inquiry question and does not necessarily have a progressive relationship, thereby causing a lower pushing question hit rate and a poor user experience.

[0005] The above information disclosed in the background section is only for the purpose of enhancing the understanding of the background of the inventive concepts, and therefore, it can contain information that does not form the prior art known to those skilled in the art in the country. SUMMARY

[0006] The summary section is provided to introduce the concepts briefly in a simplified form, which will be described in detail in the specific embodiments section later. The summary section is not intended to identify key or essential features of the claimed technical solutions nor is it intended to be used to limit the scope of the claimed technical solutions.

[0007] Some embodiments of the present disclosure propose a dialogue pushing method and device applied to a target scene, equipment and a readable medium, to solve one or more of the technical problems mentioned in the background section.

[0008] In a first aspect, some embodiments of the present disclosure provide a question pushing method applied to a target scenario, the method comprising: in response to receiving user dialogue information sent by a user terminal in a target scenario, performing text preprocessing on the user dialogue information to obtain processed dialogue information, wherein the target scenario is a customer service dialogue scenario of a user of the user terminal for a virtual item or a preset event, and the user terminal is a client that has logged in a user account; performing entity and intent recognition on the processed dialogue information according to a pre-trained entity and intent recognition model to generate dialogue entity and intent information; generating dialogue strategy information according to the dialogue entity and intent information, wherein the dialogue strategy information represents a predicted dialogue action at a next time; selecting dialogue data as target dialogue data from a pre-constructed dialogue database according to the dialogue entity and intent information and the dialogue strategy information, wherein the dialogue database comprises a set of question and answer information, a set of dialogue files, and a set of associated question information; generating a sequence of recommended question text information according to the determined target dialogue data; and pushing recommended questions to the user terminal according to the generated sequence of recommended question text information.

[0009] In a second aspect, some embodiments of the present disclosure provide a question pushing device applied to a target scenario, the device comprising: a text preprocessing unit configured to, in response to receiving user dialogue information sent by a user terminal in a target scenario, perform text preprocessing on the user dialogue information to obtain processed dialogue information, wherein the target scenario is a customer service dialogue scenario of a user of the user terminal for a virtual item or a preset event, and the user terminal is a client that has logged in a user account; an entity and intent recognition unit configured to perform entity and intent recognition on the processed dialogue information according to a pre-trained entity and intent recognition model to generate dialogue entity and intent information; a first generation unit configured to generate dialogue strategy information according to the dialogue entity and intent information, wherein the dialogue strategy information represents a predicted dialogue action at a next time; a selection unit configured to select dialogue data as target dialogue data from a pre-constructed dialogue database according to the dialogue entity and intent information and the dialogue strategy information, wherein the dialogue database comprises a set of question and answer information, a set of dialogue files, and a set of associated question information; a second generation unit configured to generate a sequence of recommended question text information according to the determined target dialogue data; and a pushing unit configured to push recommended questions to the user terminal according to the generated sequence of recommended question text information.

[0010] In a third aspect, some embodiments of the present disclosure provide an electronic device, comprising: one or more processors; a storage device having stored thereon one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement the method described in any implementation manner of the first aspect.

[0011] In a fourth aspect, some embodiments of the present disclosure provide a computer-readable medium having stored thereon a computer program, wherein the program, when executed by a processor, implements the method described in any implementation manner of the first aspect.

[0012] The above various embodiments of the present disclosure have the following beneficial effects: the dialog pushing method applied to the target scene by some embodiments of the present disclosure can improve the hitting rate of the pushed question, thereby improving the user customer service dialog experience and improving the credibility of the pushed question. Specifically, the reasons for the low hitting rate of the pushed question and the low credibility of the pushed question are as follows: first, when using a traditional dialog database for question matching, the matched pushed question only has a high semantic similarity with the user inquiry question and does not necessarily have a progressive relationship, thereby causing a low hitting rate of the pushed question and a poor user experience. Second, using a large language model to generate a pushed question in real time has a model illusion problem, thereby causing the generated pushed question to have errors, reducing the user dialog experience and system security. Based on this, the dialog pushing method applied to the target scene by some embodiments of the present disclosure first, in response to receiving user dialog information sent by a user terminal in a target scene, performs text preprocessing on the above user dialog information to obtain processed dialog information. The target scene is a customer service dialog scene of a user of the user terminal for a virtual item or a preset event. The user terminal is a client that has logged in a user account. Thus, text preprocessing can reduce noise (irrelevant characters, stop words, etc.) in the user dialog text. Then, according to a pre-trained entity intent recognition model, entity intent information of the processed dialog information is recognized to generate dialog entity intent information. The dialog entity intent information includes dialog entity information and dialog intent information. Thus, by using the entity intent recognition model, the processed dialog information can be decomposed into dialog entity information (such as a specific virtual item or target event that the user wants to inquire about) and dialog intent information (such as consultation, inquiry, complaint, etc.), thereby representing the actual dialog target of the client user through the dialog entity information and the dialog intent information. Unlike traditional matching based only on semantic similarity, intent recognition of the processed text information can predict the directionality of the dialog, so that the subsequently generated pushed question is more in line with the current needs of the user, thereby increasing the hitting rate of the pushed question. Then, according to the dialog entity information and the dialog intent information, dialog strategy information is generated. The dialog strategy information represents the predicted dialog action at the next moment. Thus, by predicting the dialog behavior based on the current dialog state of the user, the pushed question can be made more hierarchical and progressive in relation to the predicted dialog action of the user at the next moment, thereby improving the hitting rate of the pushed question. Next, according to the dialog entity information, the dialog intent information, and the dialog strategy information, dialog data is selected as target dialog data from a pre-constructed dialog database. The dialog database includes a set of question and answer information, a set of dialog files, and a set of associated question information. Thus, by filtering dialog data based on entities, intents, and dialog strategies, the selected target dialog data can have a high semantic similarity and also have a progressive relationship.Secondly, according to the determined target dialogue data, a recommended question text information sequence is generated. Finally, according to the generated recommended question text information sequence, a recommended question is pushed to the above-mentioned user end. In this way, the illusion of a large language model can be reduced by pre-set correct dialogue data, and the correctness and reliability of the generated push question can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0013] The above and other features, advantages and aspects of embodiments of the present disclosure will become more apparent by describing in detail some embodiments thereof with reference to the attached drawings in which:

[0014] Figure 1 is a flowchart of some embodiments of a question pushing method applied to a target scenario according to the present disclosure;

[0015] Figure 2 is a structural schematic diagram of some embodiments of a question pushing device applied to a target scenario according to the present disclosure;

[0016] Figure 3 is a structural schematic diagram of an electronic device suitable for implementing some embodiments of the present disclosure;

[0017] Figure 4 is a schematic internal test screenshot of a question pushing interface of a question pushing method for a target scenario according to the present disclosure. DETAILED DESCRIPTION

[0018] Embodiments of the present disclosure will be described below in more detail with reference to the accompanying drawings. Although some embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms, and should not be interpreted as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to make the present disclosure more thorough and complete. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes, and are not intended to limit the scope of protection of the present disclosure.

[0019] In addition, it should be further noted that only parts related to the invention are shown in the drawings for ease of description. The embodiments in the present disclosure and the features in the embodiments can be combined with each other without conflict.

[0020] It should be noted that the concepts of "first", "second", etc. mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not intended to limit the order or interdependence of the functions performed by these devices, modules or units.

[0021] It should be noted that the modification of "one", "multiple" mentioned in the present disclosure is illustrative but not restrictive, and those skilled in the art should understand that unless the context clearly indicates otherwise, it should be understood as "one or more".

[0022] The names of the messages or information exchanged between the plurality of devices in the embodiments of the present disclosure are only for illustrative purposes, and are not used to limit the scope of the messages or information.

[0023] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0024] Figure 1 Flow 100 of some embodiments of the problem pushing method applied to the target scene according to the present disclosure is shown. The problem pushing method applied to the target scene includes the following steps:

[0025] Step 101, in response to receiving user conversation information sent by the user end in the target scene, text pre-processing is performed on the user conversation information to obtain processed conversation information.

[0026] In some embodiments, the execution subject (such as a computing device) of the problem pushing method applied to the target scene can perform text pre-processing on the user conversation information sent by the user end in the target scene in response to receiving the user conversation information through a wired connection or a wireless connection to obtain processed conversation information. Wherein, the target scene can be a customer service conversation scene of a user of the user end for a virtual item (virtual commodity) or a preset event. For example, the virtual item can be but not limited to stocks or futures. The preset event can be a user operation event based on a target application. The target application can be an applet or an APP. For example, the target application can be a mobile banking APP, and the user operation event can be but not limited to opening a new account, closing an account, or opening a related application function. The user end can be a client that has logged in a user account. The user conversation information can be text information sent by the user on the user end.

[0027] It should be noted that the wireless connection can include but not limited to 3G / 4G / 5G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultrawideband) connection, and other now known or future developed wireless connection.

[0028] In some optional implementations of some embodiments, the execution subject can perform text pre-processing on the user conversation information to obtain processed conversation information by the following steps:

[0029] In a first step, the user dialogue information is labeled to obtain labeled user dialogue information. In practice, the execution subject can perform CLS labeling on the user dialogue information to obtain the labeled user dialogue information. For example, the execution subject can label [CLS] and [SEP] at the head and tail of the user dialogue information, respectively.

[0030] In a second step, the labeled user dialogue information is segmented according to a preset punctuation character set to obtain dialogue sentence information. The preset punctuation character set can include, but is not limited to, the following punctuation characters: semicolon (;), period (.), question mark (?), and colon (:). In practice, the execution subject can segment the labeled user dialogue information according to the preset punctuation characters in the preset punctuation character set to obtain the dialogue sentence information.

[0031] In a third step, the dialogue sentence information is segmented to obtain a dialogue sentence segmentation sequence. Each dialogue sentence segmentation in the dialogue sentence segmentation sequence corresponds to a segmentation identifier. In practice, the execution subject can segment the dialogue sentence information by using a natural language processing tool to obtain the dialogue sentence segmentation sequence. As an example, the natural language processing tool can be Jieba or Yaha. For example, the user dialogue information can be "I want to consult what is futures? Do I need to open an account separately?", the dialogue sentence information obtained after processing can be "[SEP] I want to consult what is futures, do I need to open an account separately [SEP]", and the segmentation sequence obtained after segmenting the dialogue sentence information can be ([SEP] I want / consult / what / is / futures / do / separately / open an account[SEP]).

[0032] In a fourth step, the dialogue sentence segmentation sequence is determined as the processed dialogue information.

[0033] In step 102, the processed dialogue information is subjected to entity and intent recognition according to a pre-trained entity and intent recognition model to generate dialogue entity and intent information.

[0034] In some embodiments, the execution subject can perform entity and intent recognition on the processed dialogue information according to a pre-trained entity and intent recognition model to generate dialogue entity and intent information. The dialogue entity and intent information can include dialogue entity information and dialogue intent information. The dialogue entity information can include at least one dialogue entity and a corresponding confidence level. The dialogue entity can refer to an entity in the user dialogue text. For example, the dialogue entity can be "futures" and "account". The dialogue intent information can be a label representing the user's intent to perform customer service dialogue at the user end. For example, the dialogue intent information can represent, but is not limited to, dialogue intents such as "consultation", "inquiry", or "complaint".

[0035] In some optional implementations of some embodiments, the execution subject can perform entity and intent recognition on the processed dialogue information according to a pre-trained entity and intent recognition model to generate dialogue entity and intent information by the following steps:

[0036] First, perform feature extraction on the processed dialogue information to generate dialogue feature information. The dialogue feature information can include a word sequence feature vector and a dialogue text feature vector. The word sequence feature vector can be a feature vector obtained by performing feature extraction on the word sequence of the dialogue sentence. The dialogue text feature vector can be a feature vector obtained by performing feature extraction on the labeled user dialogue information.

[0037] Second, input the word sequence feature vector and the processed dialogue information into a dialogue entity recognition model included in the entity and intent recognition model to obtain dialogue entity information. The entity and intent recognition model can include a dialogue entity recognition model and a dialogue intent classification model. The entity and intent recognition model can be a neural network model with the word sequence feature vector, the dialogue text feature vector, and the processed dialogue information as input and the dialogue entity information and the dialogue intent information as output. As an example, the dialogue entity recognition model can be composed of a Transformer neural network model using relative position encoding and a Bi-LSTM model.

[0038] Third, input the dialogue entity information, the dialogue text feature vector, and the processed dialogue information into the dialogue intent classification model to generate dialogue intent information. The dialogue intent classification model can be a classification model with the dialogue entity information, the dialogue text feature vector, and the processed dialogue information as input and the dialogue intent information as output. As an example, the dialogue intent classification model can be composed of a GRU embedding layer, a Transformer neural network, and an output layer sequentially linked.

[0039] In the fourth step, the generated dialogue entity information and dialogue intent information are determined as dialogue entity intent information.

[0040] In the process of adopting technical solutions to solve the problems mentioned in the background, the following problems are often accompanied: usually, there is a relationship between entities and intents in a dialogue. Adopting a mutually independent way to identify entities and intents in user dialogue text often loses a lot of semantic information, resulting in entity and intent recognition errors, and further leading to errors in the generated push questions.

[0041] In the face of the above technical problems, the inventors decided to adopt the following solutions:

[0042] In some optional implementations of some embodiments, the above execution subject can perform feature extraction processing on the above processed dialogue information to generate dialogue feature information by the following steps:

[0043] In the first step, for each dialogue sentence included in the above processed dialogue information, the following feature extraction steps are performed:

[0044] In the first sub-step, the first text encoding processing is performed on the above dialogue sentence segmentation to generate a first segmentation encoding feature vector. In practice, the above execution subject can perform first text encoding processing on the above dialogue sentence segmentation by a first encoding method to generate a first segmentation encoding feature vector. As an example, the above first encoding method can be a one-hot encoding method.

[0045] In the second sub-step, the second text encoding processing is performed on the above dialogue sentence segmentation to generate a second segmentation encoding feature vector. The dimensions of the above first segmentation encoding feature vector and the above second segmentation encoding feature vector are different. In practice, the above execution subject performs second text encoding processing on the above dialogue sentence segmentation by a second encoding method to generate a second segmentation encoding feature vector. As an example, the above second encoding method can be an N-gram encoding method (n≤4).

[0046] In the third sub-step, semantic embedding processing is performed on the above dialogue sentence segmentation to generate a segmentation semantic feature vector. In practice, the above execution subject can perform semantic embedding processing on the above dialogue sentence segmentation by a semantic embedding model to generate a segmentation semantic feature vector. As an example, the above semantic embedding model can be a BERT model, a ConveRT model, or a Distil-BERT model.

[0047] A fourth sub-step is to concatenate the first segmentation feature vector and the second segmentation feature vector to obtain a concatenated segmentation feature vector. In practice, the execution subject can horizontally concatenate the first segmentation feature vector and the second segmentation feature vector to obtain the concatenated segmentation feature vector. For example, the first segmentation feature vector can be A, the second segmentation feature vector can be B, and the concatenated segmentation feature vector is (A, B).

[0048] A fifth sub-step is to perform dimension reduction processing on the concatenated segmentation feature vector to obtain a reduced segmentation feature vector. In practice, the execution subject can input the concatenated segmentation feature vector into a dropout layer (i.e., a Dropout layer) to perform dimension reduction processing on the concatenated segmentation feature vector to obtain the reduced segmentation feature vector. The dropout rate of the Dropout layer can be 0.15.

[0049] A second step is to sequentially concatenate each reduced segmentation feature vector according to the position sequence of the corresponding sentence segmentation in the sentence segmentation sequence to obtain a segmentation sequence encoding feature vector.

[0050] A third step is to sequentially concatenate each segmentation semantic feature vector according to the position sequence of the corresponding sentence segmentation in the sentence segmentation sequence to obtain a segmentation sequence semantic feature vector.

[0051] A fourth step is to concatenate the segmentation sequence encoding feature vector and the segmentation sequence semantic feature vector to obtain a reduced segmentation sequence feature vector. In practice, first, the execution subject can use a feed-forward neural network to keep the dimensions of the segmentation sequence encoding feature vector and the segmentation sequence semantic feature vector consistent. Then, the segmentation sequence encoding feature vector and the segmentation sequence semantic feature vector can be concatenated using the concat operation to obtain the reduced segmentation sequence feature vector.

[0052] A fifth step is to perform semantic embedding processing on the labeled user dialogue information to generate a dialogue text semantic feature vector. In practice, the execution subject can use the semantic embedding model to perform semantic embedding processing on the labeled user dialogue information to generate the dialogue text semantic feature vector.

[0053] A sixth step is to perform text encoding processing on the labeled user dialogue information to generate a dialogue text encoding feature vector. The text encoding processing includes the first encoding processing, the second encoding processing, and feature concatenation processing. It should be noted that the implementation of generating the dialogue text encoding feature vector can refer to the implementation of the concatenated segmentation feature vector, which will not be described here.

[0054] In the seventh step, the dialogue text semantic feature vector is spliced with the dialogue text coding feature vector to obtain a dialogue text feature vector to be dimensionally reduced.

[0055] In the eighth step, the dialogue text feature vector to be dimensionally reduced is obtained by performing dimension reduction on the dialogue text feature vector to be dimensionally reduced and the segmented sequence feature vector to be dimensionally reduced. In practice, the execution subject can splice the dialogue text feature vector to be dimensionally reduced and the segmented sequence feature vector to be dimensionally reduced, and then input them into a preset dimension reduction neural network to obtain the dialogue text feature vector. The preset dimension reduction neural network can be composed of two feedforward neural networks, and the neuron weights of the two feedforward neural network layers are shared with each other. The activation function of the dimension reduction neural network can be a Relu function.

[0056] The first step to the eighth step as an inventive point of an embodiment of the present disclosure solves the technical problem that the entity and the intent in the dialogue usually have a connection, and the entity and the intent in the dialogue are identified in a mutually independent manner, which often loses a large amount of semantic information, thereby causing errors in entity and intent identification, and further causing errors in the generated push question. The factors that cause the push question to be mismatched and have errors are usually as follows: the entity and the intent in the dialogue usually have a connection, and the entity and the intent in the dialogue are identified in a mutually independent manner, which often loses a large amount of semantic information, thereby causing errors in entity and intent identification. Through the two text coding methods and the feature extraction method combining local features and global features, more fine-grained deep feature representations can be extracted, a large number of complex semantic features in Chinese sentences can be accurately captured, semantic information loss is reduced, and the identification accuracy of the intent and the entity in the sentence is further improved.

[0057] In step 103, dialogue strategy information is generated according to the dialogue entity information and the dialogue intent information.

[0058] In some embodiments, the execution subject can generate dialogue strategy information according to the dialogue entity information and the dialogue intent information. The dialogue strategy information can represent the predicted dialogue action at the next moment.

[0059] In some optional implementations of some embodiments, the execution subject can generate dialogue strategy information according to the dialogue entity information and the dialogue intent information by the following steps:

[0060] In the first step, the dialogue entity information and the dialogue intent information are respectively processed to generate dialogue entity feature vectors and dialogue intent feature vectors. In practice, the execution subject can extract semantic features from the dialogue entity information and the dialogue intent information by using the semantic embedding model to generate dialogue entity feature vectors and dialogue intent feature vectors.

[0061] Secondly, the dialogue entity feature vector and the dialogue intent feature vector are concatenated to obtain a dialogue feature vector matrix.

[0062] Thirdly, the dialogue feature vector is input into a pre-trained dialogue strategy prediction model to generate dialogue strategy information. The dialogue strategy prediction model includes a semantic enhancement network and a dialogue strategy prediction network. The semantic enhancement network can be a neural network model for enhancing semantic features. The dialogue strategy prediction model takes the dialogue feature vector as input and outputs dialogue strategy information. As an example, the dialogue strategy prediction model can be composed of an input layer, a GRU embedding layer, a Transformer network, and an output layer connected in sequence. The semantic enhancement network can be an MLM (Masked Language Model) model.

[0063] In step 104, dialogue data is selected as target dialogue data from a pre-constructed dialogue database according to the dialogue entity information, the dialogue intent information, and the dialogue strategy information.

[0064] In some embodiments, the execution subject can select dialogue data as target dialogue data from a pre-constructed dialogue database according to the dialogue entity information, the dialogue intent information, and the dialogue strategy information. The dialogue database includes a set of question and answer pair information, a set of dialogue files, and a set of associated question information. The target dialogue data can be related data for generating push question information.

[0065] In the process of using the technical solutions to solve the problems mentioned in the background, the following problems often occur: using a large language model to generate push questions in real time has a model illusion problem, which leads to errors in the generated push questions, reducing user dialogue experience and system security.

[0066] To solve the above technical problems, the inventors have decided to use the following solutions:

[0067] In some optional implementations of some embodiments, the execution subject can select dialogue data as target dialogue data from a pre-constructed dialogue database according to the dialogue entity information, the dialogue intent information, and the dialogue strategy information by the following steps:

[0068] Firstly, the dialogue strategy information is subjected to semantic embedding processing to generate a dialogue strategy semantic feature vector. In practice, the execution subject can subject the dialogue strategy information to semantic embedding processing by the semantic embedding model to generate a dialogue strategy semantic feature vector.

[0069] Secondly, according to the generated dialogue entity feature vector, the dialogue intention feature vector and the dialogue strategy semantic feature vector, each question and answer pair information satisfying the semantic matching condition is selected from the question and answer pair information set included in the dialogue database as each target question and answer pair information. Each question and answer pair information in the question and answer pair information set includes question text information and answer text information, and the question text information corresponds to the associated question information in the associated question information set. In practice, the execution subject can determine the average value of the semantic similarity between the question text information included in each question and answer pair information and the dialogue entity feature vector, the dialogue intention feature vector and the dialogue strategy semantic feature vector. The semantic matching condition can be that the average value of the semantic similarity is greater than or equal to a preset similarity threshold. As an example, the preset similarity threshold can be 85%.

[0070] Thirdly, according to the determined each target question and answer pair information, each corresponding associated question information is determined from the associated question information set as a candidate associated question information set. The associated question information includes question text information and each associated question text information. The corresponding associated question information can refer to the associated question information including the same question text information as the question text information included in the target question and answer pair information.

[0071] Fourthly, according to the generated dialogue entity feature vector, the dialogue intention feature vector and the dialogue strategy semantic feature vector, each candidate associated question information satisfying the semantic matching condition is selected from the candidate associated question information set as a target associated question information.

[0072] Fifthly, according to the generated dialogue entity feature vector, the dialogue intention feature vector and the dialogue strategy semantic feature vector, each dialogue file satisfying the file matching condition is selected from the dialogue file set as a target dialogue file. The file matching condition can be that the average value of the semantic similarity between the file name of the target dialogue file and the dialogue entity feature vector, the dialogue intention feature vector and the dialogue strategy semantic feature vector is greater than or equal to the preset similarity threshold. The semantic similarity can be determined by determining the cosine similarity between the vectors.

[0073] Sixthly, the determined target question and answer pair information, target dialogue file and target associated question information are determined as target dialogue data.

[0074] The first step to the sixth step above is an application point of an embodiment of the present disclosure, which solves the technical problem of "using a large language model to generate a push question in real time, causing model hallucination of the push question, resulting in errors in the generated push question, and reducing user conversation experience and system security." Factors that lead to a decline in user conversation experience and system security are often as follows: using a large language model to generate a push question in real time, causing model hallucination of the push question, resulting in errors in the generated push question. Thus, by extracting the dialogue entity feature vector, the dialogue intent feature vector, and the dialogue strategy semantic feature vector, the target question and answer pair information, the target dialogue file, and the target associated question information with higher semantic similarity to the dialogue text can be filtered from the dialogue database. This allows for more comprehensive and accurate background information when generating recommended questions, resulting in recommended questions that are more relevant to the user's dialogue text, thereby improving user conversation experience and system security.

[0075] Step 105, generating a recommended question text information sequence according to the determined target dialogue data.

[0076] In some embodiments, the execution subject can generate a recommended question text information sequence according to the determined target dialogue data.

[0077] In some optional implementations of some embodiments, the execution subject can generate a recommended question text information sequence according to the selected target dialogue data by the following steps:

[0078] First, input the generated target dialogue data and the preset prompt word into the recommended question generation model to generate each recommended question text information. In practice, first, the execution subject can input the target dialogue data as background information into the recommended question generation model along with the preset prompt word to generate each recommended question text information.

[0079] Second, determine each generated recommended question text information as a recommended question text information sequence. In practice, the execution subject can determine the average of the semantic similarity of each recommended question text information to the dialogue entity feature vector, the dialogue intent feature vector, and the dialogue strategy semantic feature vector. The semantic similarity can be cosine similarity. Then, the execution subject can sort the generated each recommended question text information in descending order of semantic similarity to obtain a recommended question text information sequence. As an example, the recommended question generation model can be a generative language model, i.e., a large language model.

[0080] Step 106, pushing a recommended question to the user end according to the generated recommended question text information sequence.

[0081] In some embodiments, the execution subject can push the recommended question to the user end according to the generated recommended question text information sequence. In practice, the execution subject can sequentially select the first preset number of recommended question text information from the recommended question text information sequence as the recommended question, and push the recommended question to the user end in the form of a list.

[0082] The above various embodiments of the present disclosure have the following beneficial effects: the dialog pushing method applied to the target scene by some embodiments of the present disclosure can improve the hitting rate of the pushed question, thereby improving the user customer service dialog experience and improving the credibility of the pushed question. Specifically, the reasons for the low hitting rate of the pushed question and the low credibility of the pushed question are as follows: first, when using a traditional dialog database for question matching, the matched pushed question only has a high semantic similarity with the user inquiry question and does not necessarily have a progressive relationship, thereby causing a low hitting rate of the pushed question and a poor user experience. Second, using a large language model to generate a pushed question in real time has a model illusion problem, thereby causing the generated pushed question to have errors, reducing the user dialog experience and system security. Based on this, the dialog pushing method applied to the target scene by some embodiments of the present disclosure first, in response to receiving user dialog information sent by a user terminal in a target scene, performs text preprocessing on the above user dialog information to obtain processed dialog information. The target scene is a customer service dialog scene of a user of the user terminal for a virtual item or a preset event. The user terminal is a client that has logged in a user account. Thus, text preprocessing can reduce noise (irrelevant characters, stop words, etc.) in the user dialog text. Then, according to a pre-trained entity intent recognition model, entity intent information of the processed dialog information is recognized to generate dialog entity intent information. The dialog entity intent information includes dialog entity information and dialog intent information. Thus, by using the entity intent recognition model, the processed dialog information can be decomposed into dialog entity information (such as a specific virtual item or target event that the user wants to inquire about) and dialog intent information (such as consultation, inquiry, complaint, etc.), thereby representing the actual dialog target of the client user through the dialog entity information and the dialog intent information. Unlike traditional matching based only on semantic similarity, intent recognition of the processed text information can predict the directionality of the dialog, so that the subsequently generated pushed question is more in line with the current needs of the user, thereby increasing the hitting rate of the pushed question. Then, according to the dialog entity information and the dialog intent information, dialog strategy information is generated. The dialog strategy information represents the predicted dialog action at the next moment. Thus, by predicting the dialog behavior based on the current dialog state of the user, the pushed question can be made more hierarchical and progressive in relation to the predicted dialog action of the user at the next moment, thereby improving the hitting rate of the pushed question. Next, according to the dialog entity information, the dialog intent information, and the dialog strategy information, dialog data is selected as target dialog data from a pre-constructed dialog database. The dialog database includes a set of question and answer information, a set of dialog files, and a set of associated question information. Thus, by filtering dialog data based on entities, intents, and dialog strategies, the selected target dialog data can have a high semantic similarity and also have a progressive relationship.Secondly, according to the determined target dialogue data, a recommended question text information sequence is generated. Finally, according to the generated recommended question text information sequence, a recommended question is pushed to the above-mentioned user end. In this way, the illusion of a large language model can be reduced by pre-setting correct dialogue data, and the correctness and credibility of the generated push question can be improved.

[0083] Further reference Figure 2 , as an implementation of the method shown in the above figures, the present disclosure provides some embodiments of a question pushing device applied to a target scene, which device embodiments correspond to those method embodiments shown in Figure 1 , the question pushing device applied to a target scene can be applied to various electronic devices.

[0084] As shown in Figure 2 , the question pushing device 200 applied to a target scene of some embodiments includes a text preprocessing unit 201, an entity and intent recognition unit 202, a first generation unit 203, a selection unit 204, a second generation unit 205, and a pushing unit 206. Wherein, the text preprocessing unit 201 is configured to, in response to receiving user dialogue information sent by a user end in a target scene, perform text preprocessing on the above-mentioned user dialogue information to obtain processed dialogue information, wherein the above-mentioned target scene is a customer service dialogue scene of a user of the above-mentioned user end for a virtual item or a preset event, and the above-mentioned user end is a client that has logged in a user account. The entity and intent recognition unit 202 is configured to perform entity and intent recognition on the above-mentioned processed dialogue information according to a pre-trained entity and intent recognition model to generate dialogue entity and intent information, wherein the above-mentioned dialogue entity and intent information includes dialogue entity information and dialogue intent information. The first generation unit 203 is configured to generate dialogue strategy information according to the above-mentioned dialogue entity information and the above-mentioned dialogue intent information, wherein the above-mentioned dialogue strategy information represents a predicted dialogue action at the next moment. The selection unit 204 is configured to select dialogue data as target dialogue data from a pre-constructed dialogue database according to the above-mentioned dialogue entity information, the above-mentioned dialogue intent information, and the above-mentioned dialogue strategy information, wherein the above-mentioned dialogue database includes a set of question and answer information, a set of dialogue files, and a set of associated question information. The second generation unit 205 is configured to generate a recommended question text information sequence according to the determined target dialogue data. The pushing unit 206 is configured to push a recommended question to the above-mentioned user end according to the generated recommended question text information sequence.

[0085] It can be understood that the units described in the question pushing device 200 applied to a target scene correspond to the units described with reference to Figure 1The various steps in the described methods correspond. Thus, the operations, features, and benefits described above for the methods also apply to the problem pushing apparatus 200 and units contained therein, where applicable, and are not repeated here.

[0086] Reference is made below to Figure 3 which shows a structural schematic diagram of an electronic device 300 suitable for use in implementing some embodiments of the present disclosure. Figure 3 The electronic device shown is merely an example and should not bring any limitation to the function and scope of use of embodiments of the present disclosure.

[0087] As Figure 3 shown, the electronic device 300 can include a processing apparatus (e.g., a central processor, a graphics processor, etc.) 301, which can perform various appropriate actions and processes according to programs stored in a read-only memory 302 or loaded into a random access memory 303 from a storage apparatus 308. Various programs and data required for the operation of the electronic device 300 are also stored in the random access memory 303. The processing apparatus 301, the read-only memory 302, and the random access memory 303 are connected to each other through a bus 304. An input / output interface 305 is also connected to the bus 304.

[0088] Generally, the following apparatuses can be connected to the I / O interface 305: input apparatuses 306 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc. Output apparatuses 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc. Storage apparatuses 308 including, for example, a magnetic tape, a hard disk, etc. And communication apparatuses 309. The communication apparatuses 309 can allow the electronic device 300 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 3 The electronic device 300 is shown with various apparatuses, but it should be understood that all the apparatuses shown are not required, but instead a greater number of apparatuses or fewer number of apparatuses can be implemented. Figure 3 Each block shown in the figure can represent an apparatus or a plurality of apparatuses as needed.

[0089] In particular, the processes described above with reference to the flowcharts can be implemented as a computer software program according to some embodiments of the present disclosure. For example, some embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer readable medium, the computer program comprising program code for performing the methods illustrated by the flowcharts. In some such embodiments, the computer program can be downloaded and installed from a network via the communication device 309, or installed from the storage device 308, or installed from the read only memory 302. When the computer program is executed by the processing device 301, the above-mentioned functions defined in the methods of some embodiments of the present disclosure are performed.

[0090] It should be noted that the computer readable medium in some embodiments of the present disclosure can be a computer readable signal medium or a computer readable storage medium or any combination thereof. The computer readable storage medium can be, for example but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any suitable combination of the above. More specific examples of the computer readable storage medium can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In some embodiments of the present disclosure, the computer readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus or device. In some embodiments of the present disclosure, the computer readable signal medium can include a data signal carried in a baseband or as part of a carrier wave, in which the computer readable program code is carried. Such a propagated data signal can take any of a variety of forms, including but not limited to electro-magnetic, optical, or any suitable combination thereof. The computer readable signal medium can also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate or transport a program for use by or in connection with an instruction execution system, apparatus or device. Program code contained in a computer readable medium can be transmitted using any suitable medium, including but not limited to wire, cable, RF (radio frequency), etc., or any suitable combination of the above.

[0091] In some embodiments, the client, server can communicate using any currently known or future developed network protocols, such as HTTP (HyperText Transfer Protocol), and can be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include local area networks ("LAN"), wide area networks ("WAN"), internetworks (e.g., the Internet), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future developed networks.

[0092] The computer readable medium described above can be included in the electronic device described above. It can also exist separately and not be assembled into the electronic device. The computer readable medium described above carries one or more programs, when the one or more programs are executed by the electronic device, the electronic device is caused to: in response to receiving user conversation information sent by a user terminal in a target scenario, text pre-process the user conversation information to obtain processed conversation information, wherein the target scenario is a customer service conversation scenario of the user of the user terminal for a virtual item or a preset event, and the user terminal is a client that has logged in a user account. According to a pre-trained entity-intent recognition model, entity-intent recognition is performed on the processed conversation information to generate conversation entity-intent information, wherein the conversation entity-intent information includes conversation entity information and conversation intent information. According to the conversation entity information and the conversation intent information, conversation strategy information is generated, wherein the conversation strategy information represents a predicted conversation action at the next moment. According to the conversation entity information, the conversation intent information and the conversation strategy information, conversation data is selected as target conversation data from a pre-constructed conversation database, wherein the conversation database includes a set of question-answer pair information, a set of conversation files and a set of associated question information. According to the determined target conversation data, a sequence of recommended question text information is generated. According to the generated sequence of recommended question text information, a recommended question is pushed to the user terminal.

[0093] Computer program code for carrying out operations of some embodiments of the present disclosure can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like, and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0094] The flow diagrams and the block diagrams in the drawings are illustrations of architectures, functionalities, and operations of possible implementations of systems, methods, and computer program products according to various embodiments of present disclosure. In this regard, each block in the flow diagrams or block diagrams can represent a module, a procedure, or a part of code, which comprises one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figures. For example, two blocks shown in succession may, in some cases, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It is also noted that each block of the block diagrams and / or flow diagrams and combinations of blocks in the block diagrams and / or flow diagrams can be implemented by a dedicated hardware-based system that carries out specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0095] The units described in some embodiments of the present disclosure can be implemented by means of software, or can be implemented by hardware. The described units can also be provided in a processor, for example, it can be described that: a processor includes a text preprocessing unit, an entity and intent recognition unit, a first generation unit, a selection unit, a second generation unit, and a pushing unit. Among them, the names of these units do not constitute a limitation to the units themselves in some cases, for example, the text preprocessing unit can also be described as: "a unit that, in response to receiving user dialogue information sent by a user end in a target scene, performs text preprocessing on the user dialogue information to obtain processed dialogue information".

[0096] The functionality described herein above can be performed, at least in part, by one or more hardware logic components. For example, and without limitation, illustrative types of hardware logic components that can be used include Field-programmable Gate Arrays (FPGAs), Program- specific Integrated Circuits (ASICs), Program- specific Standard Products (ASSPs), System-on-a-chip systems (SOCs), Complex Programmable Logic Devices (CPLDs), etc.

[0097] The above description is merely exemplary of the disclosure and the application made use of the principles of the technology. It is to be understood that the application scope of the embodiments of the disclosure is not limited to the specific combinations of technical features described above, and should also cover other technical solutions formed by any combination of the above technical features or equivalent features thereof without departing from the inventive concept. For example, the technical solutions formed by replacing the above features with technical features having similar functions disclosed in the embodiments of the disclosure (but not limited to) with each other.

Claims

1. A method for pushing questions to a target scenario, comprising: In response to receiving user dialogue information sent by the user terminal in a target scenario, the user dialogue information is preprocessed to obtain processed dialogue information. The target scenario is a customer service dialogue scenario where the user on the user terminal is using a target application to interact with virtual items or preset events. The user terminal is a client that has logged into a user account, and the user account is the application account of the target application. Based on a pre-trained entity intent recognition model, entity intent recognition is performed on the processed dialogue information to generate dialogue entity intent information, wherein the dialogue entity intent information includes dialogue entity information and dialogue intent information. Based on the dialogue entity information and the dialogue intent information, dialogue strategy information is generated, wherein the dialogue strategy information represents the predicted dialogue action of the user on the user terminal at the next moment; Based on the dialogue entity information, the dialogue intent information, and the dialogue strategy information, dialogue data is selected from a pre-built dialogue database as target dialogue data, wherein the dialogue database includes a question-answer pair information set, a dialogue file set, and a set of associated question information; Based on the selected target dialogue data, generate a sequence of recommendation question text information; Based on the generated sequence of recommended question text information, at least one corresponding recommended question is pushed to the user terminal.

2. The method according to claim 1, wherein, The step of preprocessing the user dialogue information to obtain processed dialogue information includes: The user dialogue information is annotated to obtain annotated user dialogue information; According to a preset punctuation character set, the annotated user dialogue information is processed into sentence segmentation to obtain dialogue sentence information; The dialogue statement information is segmented to obtain a dialogue statement segmentation sequence, wherein each dialogue statement segment in the dialogue statement segmentation sequence has a corresponding segmentation identifier. The word segmentation sequence of the dialogue statements is determined as the processed dialogue information.

3. The method according to claim 2, wherein, The step of performing entity intent recognition on the processed dialogue information based on a pre-trained entity intent recognition model to generate dialogue entity intent information includes: The processed dialogue information is subjected to feature extraction processing to generate dialogue feature information, wherein the dialogue feature information includes word segmentation sequence feature vector and dialogue text feature vector; The word segmentation sequence feature vector and the processed dialogue information are input into the dialogue entity recognition model included in the entity intent recognition model to obtain dialogue entity information. The entity intent recognition model includes a dialogue entity recognition model and a dialogue intent classification model. The dialogue entity information, the dialogue text feature vector, and the processed dialogue information are input into the dialogue intent classification model to generate dialogue intent information; The generated dialogue entity information and dialogue intent information are identified as dialogue entity intent information.

4. The method according to claim 3, wherein, The step of generating dialogue strategy information based on the dialogue entity information and the dialogue intent information includes: The dialogue entity information and the dialogue intent information are respectively characterized to generate dialogue entity feature vectors and dialogue intent feature vectors; The dialogue entity feature vector and the dialogue intent feature vector are concatenated to obtain a dialogue feature vector matrix. The dialogue feature vector is input into a pre-trained dialogue policy prediction model to generate dialogue policy information, wherein the dialogue policy prediction model includes a semantic enhancement network and a dialogue policy prediction network.

5. The method according to claim 4, wherein, The step of generating a sequence of recommendation question text information based on the selected target dialogue data includes: The generated target dialogue data and preset prompts are input into the recommendation question generation model to generate text information for each recommendation question; The generated recommendation question text information is determined as a recommendation question text information sequence.

6. A dialogue push device applied to a target scenario, comprising: The text preprocessing unit, in response to receiving user dialogue information sent by the user terminal in a target scenario, performs text preprocessing on the user dialogue information to obtain processed dialogue information, wherein the target scenario is a customer service dialogue scenario of the user terminal through a target application regarding virtual items or preset events, the user terminal is a client that has logged into a user account, and the user account is the application account of the target application. The entity intent recognition unit performs entity intent recognition on the processed dialogue information according to a pre-trained entity intent recognition model to generate dialogue entity intent information, wherein the dialogue entity intent information includes dialogue entity information and dialogue intent information. The first generation unit generates dialogue strategy information based on the dialogue entity information and the dialogue intent information, wherein the dialogue strategy information represents the predicted dialogue action of the user on the user terminal at the next moment. The selection unit selects dialogue data as target dialogue data from a pre-built dialogue database based on the dialogue entity information, the dialogue intent information, and the dialogue strategy information. The dialogue database includes a question-answer pair information set, a dialogue file set, and a set of associated question information. The second generation unit generates a sequence of recommendation question text information based on the selected target dialogue data; The push unit pushes at least one corresponding recommendation question to the user terminal based on the generated sequence of recommendation question text information.

7. An electronic device, comprising: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 5.

8. A computer-readable medium having a computer program stored thereon, wherein, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 5.

Images