Dialogue text information extraction method and device, equipment and medium

Abstract

The invention provides a dialogue text information extraction method and device, equipment and a medium, and relates to the technical field of network installation and maintenance. The method comprises the following steps: dialogue texts of a calling user and a called user are obtained, and the dialogue texts comprise a calling dialogue text, a called dialogue text and a complete dialogue text between the calling user and the called user; according to the dialogue text of the calling user, determining whether the dialogue text of the calling user and the called user is the dialogue text of the reserved target service; if yes, marking a complete dialogue text between the calling user and the called user according to the calling dialogue text and the called dialogue text to obtain a plurality of dialogue text pairs; and based on a pre-trained reservation information extraction model, extracting reservation information of the target service from the plurality of dialogue text pairs. According to the invention, the accuracy of obtaining the reservation information from the call record can be improved, and the service quality of installation and maintenance service personnel and the customer satisfaction can be improved.

Description

Background Art

[0002] With the rapid development of information technology and the widespread application of the Internet, broadband networks have become one of the basic infrastructures of modern society, which has had a profound impact on people's daily life, work and study. In major cities, the penetration rate and user scale of broadband networks have reached unprecedented heights. However, with the rapid growth of broadband services, the quality and efficiency of installation and maintenance services have become important factors restricting user experience and the development of operators.

[0003] In actual installation and maintenance scenarios, there will be a large number of voice calls between installation and maintenance service personnel and customers. The purpose of the calls may mainly include fault reporting, repair assistance, etc. The voice call may involve the customer's appointment time information, address information, etc. In theory, this information needs to be consistent with the time filled in by the installation and maintenance service personnel in the work order. In particular, the accuracy of the appointment time will directly affect the service quality and customer satisfaction of the installation and maintenance service personnel. Therefore, the accurate extraction of the appointment time is an important content that needs to be dealt with in the installation and maintenance scenario.

[0004] In some scenarios, manual listening to recordings is adopted. However, since the data volume of installation and maintenance call records is too large, it requires a lot of manpower and material resources. At the same time, the manual extraction process is easily affected by external factors, and recording errors or omissions may occur. In addition, the installation and maintenance call texts that need to be processed often have characteristics such as colloquialism and multiple rounds, and due to the translation through the automatic voice equipment ASR function, information loss may occur. In addition, sometimes multiple information involving time points will appear during the call, and it is even more necessary to accurately extract information such as the final valid appointment time point or time range.

[0005] In this regard, the related technology adopts the basic sequence-to-sequence S2S model to propose a solution for obtaining the description sentence of the appointment time, but the basic model is usually combined with the attention mechanism. The process in which the encoder encodes the input content to obtain a fixed-size state vector is a "lossy information compression" process. The greater the amount of information, the greater the loss of information in the process of converting the input vector. At the same time, as the sequence length increases, that is, the sequence length in the time dimension increases, the recurrent neural network RNN ​​will also experience gradient diffusion.

[0006] In some cases, both the encoding and decoding ends use a long short-term memory network (LTSM). The LSTM structure is relatively complex and has many parameters, and usually requires more steps to update its internal state. Therefore, the training time and prediction time are relatively long.

[0007] In addition, although the attention mechanism is added in the related technology to capture the expected information more accurately, the generated sentences are still likely to contain some interference words due to the influence of factors such as the quality of the training data.

[0008] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0009] The present disclosure provides a method, device, equipment and medium for extracting information from a conversation text, which at least to a certain extent overcomes the problem in the related art that target reservation information cannot be accurately extracted from a conversation text.

[0010] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.

[0011] According to one aspect of the present disclosure, a method for extracting conversation text information is provided, comprising: obtaining a conversation text between a calling user and a called user, wherein the conversation text comprises: a calling conversation text, a called conversation text and a complete conversation text between the calling user and the called user; determining, based on the conversation text of the calling user, whether the conversation text between the calling user and the called user is a conversation text for booking a target service; if so, marking the complete conversation text between the calling user and the called user based on the calling conversation text and the called conversation text to obtain a plurality of conversation text pairs, wherein each conversation text pair comprises a calling conversation text and a called conversation text; and extracting the reservation information of the target service from the plurality of conversation text pairs based on a pre-trained reservation information extraction model.

[0012] In some embodiments, the method of extracting the reservation information of the target business from the multiple dialogue text pairs based on a pre-trained reservation information extraction model includes: segmenting the multiple dialogue text pairs according to preset lengths to obtain multiple groups of dialogue text pairs, wherein each group of dialogue text pairs includes at least one dialogue text pair; inputting each group of dialogue text pairs into the pre-trained reservation information extraction model in sequence until the reservation information extraction model outputs the reservation information of the target business, and stopping inputting the segmented dialogue text pairs into the reservation information extraction model.

[0013] In some embodiments, before extracting the reservation information of the target business from the multiple dialogue text pairs based on a pre-trained reservation information extraction model, the method also includes: obtaining a preset prompt word format corresponding to a pre-trained large language model; calling the preset prompt word format to generate a prompt sentence related to the target business based on multiple dialogue text pairs; inputting the prompt sentence related to the target business into the pre-trained large language model to obtain summary information related to the target business.

[0014] In some embodiments, extracting the reservation information of the target business from the multiple dialogue text pairs based on a pre-trained reservation information extraction model includes: inputting summary information related to the target business into the pre-trained reservation information extraction model to obtain the reservation information of the target business.

[0015] In some embodiments, after extracting the reservation information of the target business from the multiple dialogue text pairs based on a pre-trained reservation information extraction model, the method further includes: matching the reservation information of the target business with a pre-constructed reservation information rule matching table to obtain the target reservation information, wherein the reservation information rule matching table is used to remove text in the reservation information of the target business that is not related to the target reservation information.

[0016] In some embodiments, the target appointment information includes an appointment time.

[0017] According to another aspect of the present disclosure, a conversation text information extraction device is also provided, including: a conversation text acquisition module, used to acquire the conversation text between the calling user and the called user, wherein the conversation text includes: the calling conversation text, the called conversation text and the complete conversation text between the calling user and the called user; a conversation text judgment module, used to determine whether the conversation text between the calling user and the called user is the conversation text for booking a target service based on the conversation text of the calling user; a conversation text reconstruction module, used to mark the complete conversation text between the calling user and the called user based on the calling conversation text and the called conversation text to obtain multiple conversation text pairs, wherein each conversation text pair includes a calling conversation text and a called conversation text; and a reservation information extraction module, used to extract the reservation information of the target service from the multiple conversation text pairs based on a pre-trained reservation information extraction model.

[0018] According to another aspect of the present disclosure, an electronic device is also provided, which includes: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute any one of the above-mentioned methods for extracting conversation text information by executing the executable instructions.

[0019] According to another aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method for extracting conversation text information described in any one of the above is implemented.

[0020] According to another aspect of the present disclosure, a computer program product is also provided, including a computer program, wherein when the computer program is executed by a processor, the computer program implements any one of the above-mentioned methods for extracting conversation text information.

[0021] The conversation text information extraction method, device, equipment and medium provided in the embodiments of the present disclosure, after obtaining the conversation text between the calling user and the called user, determine whether the obtained conversation text is the conversation text of the reservation target service according to the conversation text of the calling user, and if so, mark the complete conversation text between the calling user and the called user according to the obtained conversation text to obtain multiple conversation text pairs, and then extract the reservation information of the target service from the multiple conversation text pairs based on the pre-trained reservation information extraction model. The embodiments of the present disclosure can improve the accuracy of obtaining reservation information from call recordings, thereby improving the service quality and customer satisfaction of installation and maintenance service personnel.

[0022] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0024] Figure 1 A schematic diagram of a model structure used in related technologies is shown;

[0025] Figure 2 A schematic diagram of a model structure used in an embodiment of the present disclosure is shown;

[0026] Figure 3 A flow chart of a method for extracting conversation text information in an embodiment of the present disclosure is shown;

[0027] Figure 4 Another flow chart of a method for extracting conversation text information in an embodiment of the present disclosure is shown;

[0028] Figure 5 Another flow chart of a method for extracting conversation text information in an embodiment of the present disclosure is shown;

[0029] Figure 6 A schematic diagram of a conversation text information extraction device in an embodiment of the present disclosure is shown;

[0030] Figure 7 A structural block diagram of an electronic device in an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0031] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the disclosure will be more comprehensive and complete and to fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0032] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0033] For ease of understanding, before introducing the embodiments of the present disclosure, several terms involved in the embodiments of the present disclosure are first explained as follows:

[0034] encoder: encoder.

[0035] decoder: decoder.

[0036] ASR: Automatic Speech Recognition.

[0037] RNN: Recurrent Neural Network.

[0038] Bi-RNN: Bidirectional Recurrent Neural Network, bidirectional recurrent neural network.

[0039] LSTM: Long Short-Term Memory, long short-term memory network.

[0040] Bi-LSTM: Bidirectional Long Short-Term Memory, bidirectional long short-term memory network.

[0041] S2S model: Sequence-to-Sequence, sequence to sequence model.

[0042] GRU: Gated Recurrent Unit, gated recurrent unit.

[0043] Bi-GRU: Bidirectional Gated Recurrent Unit, bidirectional gated recurrent unit.

[0044] prompt: prompt word.

[0045] The specific implementation of the embodiment of the present disclosure is described in detail below with reference to the accompanying drawings.

[0046] Figure 1 FIG. 1 shows a schematic diagram of a model structure used in related technologies. Figure 1 As shown, the model may include: an encoding end 101 and a decoding end 102 .

[0047] In one embodiment of the present disclosure, there are first input, second input and third input on the encoding end 101 side, which are the time steps of the input sequence (i.e., the input at each moment), and the input of each time step can be a word, a character or other types of sequence elements.

[0048] In one embodiment of the present disclosure, the encoding end 101 also has an encoder, which is composed of one or more long short-term memory networks LSTM or recurrent neural networks RNN. The input of each time step is processed by these networks to generate a hidden state sequence.

[0049] In one embodiment of the present disclosure, there is a decoder on the decoding end 102, and the decoder is composed of one or more long short-term memory networks LSTM or recurrent neural networks RNN. The input of the decoder is usually the output of the previous moment and the hidden state generated by the encoder.

[0050] In one embodiment of the present disclosure, the attention mechanism is used on the decoding end 102 side, and the attention mechanism allows the decoder to focus on different parts of the input sequence when generating each output. Specifically, the attention mechanism generates a weight distribution based on the current state of the decoder and the hidden state of the encoder, and the weight distribution is used to weight the hidden state of the encoder to generate a context vector.

[0051] In one embodiment of the present disclosure, there is a fully connected layer on the decoding end 102, and the output of the decoder is processed by a fully connected layer to generate a final output sequence. The first output, the second output, and the third output corresponding to each moment, each output can be a word, a character, or other types of sequence elements.

[0052] In one embodiment of the present disclosure, since the encoding end 101 encodes all input sequences into the same feature vector based on the sequence-to-sequence S2S model, the process of obtaining a fixed-size state vector is a process of lossy information compression, requiring that the feature vector C must contain all information in the original sequence. If the amount of input information is larger, the loss of information in the process of converting the input into a vector is greater, which limits the performance of the model. At the same time, as the sequence length increases, that is, the sequence in the time dimension becomes longer, the RNN model will also experience gradient diffusion.

[0053] In one embodiment of the present disclosure, if both the encoding end 101 and the decoding end 102 use LSTM, at this time, since the LSTM structure is relatively complex and has more parameters, it is usually necessary to perform cumbersome steps to update its internal state, so the training time and prediction time will be relatively long.

[0054] In one embodiment of the present disclosure, although the decoding end 102 adopts the attention mechanism to more accurately capture the desired information, due to the influence of factors such as the quality of the training data, there is a possibility that some interference words may be mixed in the obtained sentence segmentation results.

[0055] For the above questions, Figure 2 FIG. 2 shows a schematic diagram of the model structure used in the embodiment of the present disclosure. Figure 2 As shown, the model may include: an encoding end 101 and a decoding end 102 .

[0056] In one embodiment of the present disclosure, there are first input, second input and third input on the encoding end 101 side, which are respective moments of the input sequence, and the input at each moment may be a word, a character or other types of sequence elements.

[0057] In one embodiment of the present disclosure, there is an encoder on the encoding end 101 side, and the encoder is composed of two LSTM layers, namely, a forward LSTM and a backward LSTM. The input of each time step is processed through these networks to generate a forward hidden state sequence and a backward hidden state sequence, respectively. Among them, the forward LSTM processes the order of the input sequence, from the first input to the third input; the backward LSTM processes the reverse order of the sequence, from the third input to the first input.

[0058] In one embodiment of the present disclosure, there is a decoder on the decoding end 102, and the decoder is composed of a plurality of gated recurrent unit (GRU) layers, and the output of each time step is generated by the GRU layer.

[0059] In one embodiment of the present disclosure, the attention mechanism is used on the decoding end 102 side, and the attention mechanism allows the decoder to focus on different parts of the input sequence when generating each output. Specifically, the attention mechanism generates a weight distribution based on the current state of the decoder and the hidden state of the encoder, and the weight distribution is used to weight the hidden state of the encoder to generate a context vector.

[0060] In one embodiment of the present disclosure, there is a fully connected layer on the decoding end 102, and the output of the decoder is processed by a fully connected layer to generate a final output sequence. The first output, the second output, and the third output corresponding to each moment, each output can be a word, a character, or other types of sequence elements.

[0061] In one embodiment of the present disclosure, the attention mechanism can obtain more useful feature information from key areas, reduce resource loss, and improve task efficiency. From the perspective of neural networks, the attention mechanism can record the positional relationship between information and obtain the correlation within the data, assign different weights to the input features of the model, and highlight more critical influencing factors. By strengthening key information and weakening useless information, the efficiency of the algorithm is improved, and the impact of input length and input order on system efficiency is reduced. The attention mechanism enables the network to focus on highly relevant information, and can directly establish a dependency relationship between input and output without a loop. It has a strong degree of parallelization and greatly improves the running speed.

[0062] In one embodiment of the present disclosure, the attention mechanism first calculates the attention probability distribution value e according to the input X. ij , the specific calculation formula is as follows:

[0063] e ij =v tanh(ω[h i ;d j ]+b) (1)

[0064] Among them, e ij represents the unnormalized attention score of the decoder at the jth moment to the encoder at the ith moment, v and ω are weight coefficients, tanh() is the activation function, and h i represents the hidden state of the encoder at the i-th moment, d j represents the hidden state of the decoder at the jth moment, [·;·] represents concatenation, and b is the bias coefficient.

[0065] In one embodiment of the present disclosure, the softmax function is used to normalize the attention score value to obtain the attention weight a ij , the specific calculation formula is as follows:

[0066]

[0067] Among them, a ij represents the normalized attention weight of the decoder at the jth moment to the encoder at the ith moment, softmax() is the activation function, e ij represents the unnormalized attention score of the decoder at the jth moment to the encoder at the ith moment, T i Indicates the total encoder processing time.

[0068] In one embodiment of the present disclosure, the hidden state h at each moment is i And the corresponding attention weight a ij Perform weighted summation to obtain the final output state value s j , the specific calculation formula is as follows:

[0069]

[0070] Among them, s j Indicates the state value of the final output (which can be regarded as context information in some cases), a ij represents the normalized attention weight of the decoder at the jth moment to the encoder at the ith moment, h i represents the hidden state of the encoder at the i-th moment, T i Indicates the total encoder processing time.

[0071] In one embodiment of the present disclosure, AI technology is used to accurately extract appointment information for scenarios such as on-site appointments and appointment changes mentioned in the installation and maintenance dialogue process, such as specific time points or time range information, thereby automatically extracting installation and maintenance appointment information, reducing manual intervention, and improving the service quality and efficiency of installation and maintenance.

[0072] Figure 3 A flow chart of a method for extracting conversation text information in an embodiment of the present disclosure is shown as follows: Figure 3 As shown, the method comprises the following steps:

[0073] S302, obtaining a conversation text between the calling user and the called user, wherein the conversation text includes: a calling conversation text, a called conversation text, and a complete conversation text between the calling user and the called user.

[0074] In one embodiment of the present disclosure, through automatic speech recognition ASR, the conversation recording between the customer and the installation and maintenance personnel can be converted into three conversation contents, which can be specifically divided into the calling conversation text caller_text, the called conversation text callee_text and the complete conversation text full_text between the calling user and the called user.

[0075] S304: Determine, based on the conversation text of the calling user, whether the conversation text between the calling user and the called user is a conversation text for a reserved target service.

[0076] In one embodiment of the present disclosure, the caller conversation text caller_text is input into a pre-trained multi-classification model, and the model outputs whether the caller conversation text involves installation and maintenance fault reporting, and if so, the specific fault type to which the fault reporting belongs.

[0077] In one embodiment of the present disclosure, the pre-trained multi-classification model can be a transformer-based bidirectional encoder representation BERT pre-trained model to implement a multi-classification model or a single classification model, a multi-task model, etc., or combined with a rule-based matching model to implement a model that can analyze whether the call text involves maintenance and troubleshooting, and when the call text involves maintenance and troubleshooting, analyze the fault type. Fault types may include but are not limited to wireless network WiFi, landline, TV, etc.

[0078] S306: If yes, then mark the complete conversation text between the calling user and the called user according to the calling conversation text and the called conversation text to obtain multiple conversation text pairs, wherein each conversation text pair includes a calling conversation text and a called conversation text.

[0079] In one embodiment of the present disclosure, the conversation content is organized into conversation pairs based on the three conversation texts obtained above, and a calling identifier caller: or a called identifier callee: is added to each corresponding conversation to mark each sentence as belonging to the calling conversation text or the called conversation text, and the reorganized conversation content is used as the conversation text content union_text in the question-and-answer mode, that is, multiple conversation text pairs are formed, each conversation text pair includes a calling conversation text and a called conversation text.

[0080] S308, extracting the reservation information of the target service from the plurality of dialogue text pairs based on the pre-trained reservation information extraction model.

[0081] In one embodiment of the present disclosure, a pre-trained reservation information extraction model is mainly composed of an encoding end, a decoding end and an extended vocabulary rule filtering module. Specifically, the model uses a bidirectional long short-term memory network Bi-LTSM as an encoding unit and a gated recurrent unit GRU as a decoding unit, and by integrating an attention mechanism, it can effectively reduce information loss.

[0082] As can be seen from the above, after obtaining the conversation text between the calling user and the called user, the embodiment of the disclosure determines whether the obtained conversation text is the conversation text of the reservation target service according to the conversation text of the calling user. If so, the complete conversation text between the calling user and the called user is marked according to the obtained conversation text to obtain multiple conversation text pairs, and then based on the pre-trained reservation information extraction model, the reservation information of the target service is extracted from the multiple conversation text pairs. The embodiment of the disclosure can improve the accuracy of obtaining reservation information from call recordings, thereby improving the service quality and customer satisfaction of installation and maintenance service personnel.

[0083] In one embodiment of the present disclosure, the above S308 includes: segmenting the multiple dialogue text pairs according to preset lengths to obtain multiple groups of dialogue text pairs, wherein each group of dialogue text pairs includes at least one dialogue text pair; inputting each group of dialogue text pairs into a pre-trained reservation information extraction model in sequence until the reservation information extraction model outputs the reservation information of the target business, and stopping inputting the segmented dialogue text pairs into the reservation information extraction model.

[0084] In one embodiment of the present disclosure, the total number of dialogue segments count in the dialogue text content union_text of the question-answer mode is counted, and the total number of dialogue segments count is segmented according to a preset length n as a unit to obtain multiple groups of dialogue text pairs. Each group of dialogue text pairs is used as a model input prompt, and all prompts are organized into a prompts list.

[0085] It should be noted that the preset length for segmenting a conversation needs to be set according to actual conditions, and the embodiments of the present disclosure do not specifically limit this.

[0086] In one embodiment of the present disclosure, when the pre-trained multi-classification model outputs a dialogue content of a customer reporting a maintenance fault, each prompt in the prompts list is traversed, and each prompt is used as an input to the appointment information extraction model. If the corresponding prompt contains a relevant description of the appointment information, the appointment information extraction model extracts the relevant appointment information and stops traversing the prompts list; if the appointment information extraction model does not extract the relevant appointment information, it continues to traverse the next prompt in the prompts list.

[0087] In one embodiment of the present disclosure, before the above S308, the method further includes: obtaining a preset prompt word format corresponding to a pre-trained large language model; calling the preset prompt word format to generate a prompt sentence related to the target business according to multiple dialogue text pairs; inputting the prompt sentence related to the target business into the pre-trained large language model to obtain summary information related to the target business.

[0088] In one embodiment of the present disclosure, the conversation text content union_text in the question-answer mode may be preprocessed to obtain a plurality of preprocessed conversation text pairs. The preprocessing method may include but is not limited to data cleaning.

[0089] In one embodiment of the present disclosure, a preset prompt word format corresponding to a pre-trained large language model is obtained, a plurality of pre-processed dialogue text pairs are processed through the preset prompt word format to generate corresponding model prompt sentences, and the prompt words are input into the pre-trained large language model to extract summary information related to installation and maintenance in the plurality of dialogue pairs to obtain summary information related to the target business, which needs to include necessary information elements related to installation and maintenance, and may include but is not limited to the appointment time for installation and maintenance, the appointment location, and the content of the services required for installation and maintenance.

[0090] In one embodiment of the present disclosure, the above S308 includes: inputting summary information related to the target service into a pre-trained reservation information extraction model to obtain reservation information of the target service.

[0091] In one embodiment of the present disclosure, after the above S308, the method further includes: matching the reservation information of the target service with a pre-constructed reservation information rule matching table to obtain the target reservation information, wherein the reservation information rule matching table is used to remove text in the reservation information of the target service that is not related to the target reservation information.

[0092] In one embodiment of the present disclosure, since the reservation information of the target business output by the above-mentioned reservation information extraction model is affected by the training effect of the model, the quality of the text, etc., interference words that do not conform to the reservation information may exist in the output result. At this time, an extended vocabulary rule filtering module is set to match the reservation information of the target business with the pre-built reservation information rule matching table to eliminate all texts that are not related to the reservation information, and then accurately obtain the target reservation information, that is, the reservation information text with the smallest text length.

[0093] In one embodiment of the present disclosure, the reservation information rule matching table can be a table constructed based on keywords used to describe the reservation information in the pre-acquired historical data. The extracted keywords are matched with the keywords in the table, and the words that fail to match can be determined to be keywords unrelated to the reservation information, and are removed from the sentence.

[0094] In one embodiment of the present disclosure, the target appointment information includes the appointment time, and may also include the appointment address, appointment time range, service type, etc. The specific content contained in the target appointment information can be determined according to actual conditions, and the embodiment of the present disclosure does not make specific limitations on this.

[0095] In one embodiment of the present disclosure, if the dialogue text obtained after translation by the ASR platform is: "1: Where is Xiao Hongying's head? 2: It's just outside the Guang Square. 1: The square, the square, um, I miss it so much. Um, I miss your household registration, uh, wait a minute, if you want to wait, then wait for something, you may have to wait. I think it's after nine o'clock now, maybe you have to wait for half an hour, maybe you can't wait until half past four. 2: Okay, okay. Okay, okay, we'll wait for you, okay.", then the key description text containing the appointment information output by the model may be: "Maybe you can't wait until half past four", and after matching with the pre-built appointment information rule matching table, the final time description obtained is "half past four".

[0096] In one embodiment of the present disclosure, the content of the conversation between the installation and maintenance personnel and the customer is first analyzed. Since the conversations of the installation and maintenance personnel are not necessarily all related to installation and maintenance, it is necessary to determine whether the conversation is related to installation and maintenance based on the key information of the conversation content. For conversations related to installation and maintenance, valid appointment information is extracted based on the conversation content, and the appointment information is provided to the installation and maintenance comprehensive scheduling system as part of the key information for generating work orders. The installation and maintenance service personnel can make reasonable arrangements based on the appointment information, thereby improving the service quality of installation and maintenance and improving customer satisfaction with the service.

[0097] Figure 4 Another flow chart of a method for extracting conversation text information in an embodiment of the present disclosure is shown as follows: Figure 4 As shown, the method comprises the following steps:

[0098] S401, obtaining a conversation text between a calling user and a called user.

[0099] In one embodiment of the present disclosure, the conversation text includes: a calling conversation text, a called conversation text, and a complete conversation text between the calling user and the called user.

[0100] S402, processing the acquired conversation text between the calling user and the called user, specifically, including the following S4021 to S4023.

[0101] S4021, organize the acquired conversation content into conversation pairs, and add a caller identifier caller: or a called identifier callee: to each corresponding conversation to mark whether each sentence belongs to the calling conversation text or the called conversation text, and use the reorganized conversation content as the conversation text content union_text in the question-and-answer mode.

[0102] S4022, counting the total number of dialogue segments count in the dialogue text content union_text of the question-answer mode, and dividing the total number of dialogue segments count according to a preset length n as a unit to obtain multiple groups of dialogue text pairs.

[0103] S4023, taking each group of dialogue text pairs as a model input prompt, and organizing all prompts into a prompts list.

[0104] S403, input the caller conversation text into the pre-trained multi-classification model to determine whether the conversation text between the calling user and the called user is the conversation text of the reservation target service. If so, execute S404; if not, continue to execute S402 based on the latest acquired conversation text.

[0105] S404: If yes, determine whether there is a prompt to be processed in the pre-obtained prompts list. If yes, execute S405; if not, continue to execute S402 based on the latest acquired dialogue text.

[0106] S405, traverse each prompt to be processed in the prompts list obtained in advance, and use each prompt as an input of the reservation information extraction model until the reservation information of the target service is obtained, and stop inputting prompts to the reservation information extraction model.

[0107] In one embodiment of the present disclosure, if there is a relevant description about the appointment information in the corresponding prompt, the appointment information extraction model extracts the relevant appointment information and stops traversing the prompts list; if the appointment information extraction model does not extract the relevant appointment information, it continues to traverse the next prompt in the prompts list.

[0108] S406, matching the reservation information of the target service with the pre-built reservation information rule matching table to obtain the target reservation information.

[0109] In one embodiment of the present disclosure, the pre-constructed reservation information rule matching table may include a custom dictionary and filtering rules, which may be a custom dictionary and filtering rules constructed based on keywords used to describe reservation information in pre-acquired historical data.

[0110] As can be seen from the above, after obtaining the conversation text between the calling user and the called user, the embodiment of the disclosure determines whether the obtained conversation text is the conversation text of the reservation target service according to the conversation text of the calling user. If so, the complete conversation text between the calling user and the called user is marked according to the obtained conversation text to obtain multiple conversation text pairs, and then based on the pre-trained reservation information extraction model, the reservation information of the target service is extracted from the multiple conversation text pairs. The embodiment of the disclosure can improve the accuracy of obtaining reservation information from call recordings, thereby improving the service quality and customer satisfaction of installation and maintenance service personnel.

[0111] In one embodiment of the present disclosure, a prompt is taken as an example. Each prompt is converted into a vector of fixed length through word embedding technology, that is, after the embedding process, a series of word vectors are obtained to form a word vector sequence. If there are n words in a sentence, n word vectors will be obtained, which can be expressed as X = {x1, x2, ..., x n}.

[0112] In one embodiment of the present disclosure, since the model uses Bi-LTSM as the encoding unit and GRU as the decoding unit, and integrates the attention mechanism, it can effectively reduce information loss. The input sequence can be expressed as X = {x1, x2, ..., x n}, the output sequence can be expressed as Y = {y1,y2,…,y n}.

[0113] In one embodiment of the present disclosure, the encoder part involves the following calculation, and the state of the hidden layer of the encoder corresponding to the input sequence can be expressed as H = {h1, h2, ..., h n}:

[0114]

[0115] in, represents the forward hidden state at time t, represents the backward hidden state at time t, represents the forward LSTM unit, represents the backward LSTM unit, x t represents the input at time t of the input sequence, represents the forward hidden state at the previous moment, represents the backward hidden state at the previous moment, Indicates splicing, h t represents the final hidden state at time t.

[0116] In one embodiment of the present disclosure, the decoder part involves the following calculation, and the state of the decoder hidden layer can be expressed as D = {d1, d2, ..., d n}:

[0117] d t =GRU dec (y t-1 ,d t-1 ) (7)

[0118] Among them, d t represents the hidden state of the decoder at time t, GRU dec () represents the GRU unit decoder, y t-1 represents the output of the decoder at the previous moment, d t-1 Represents the hidden state of the decoder at the previous moment.

[0119] In one embodiment of the present disclosure, the attention score can be calculated by the above formula (1), the normalized attention weight can be calculated by the above formula (2), and the context vector can be calculated by the above formula (3).

[0120] In one embodiment of the present disclosure, three conversation texts of call information between customers and installation and maintenance personnel are collected by the voice data collection platform. First, the conversation is reorganized into a standard conversation model according to the content of the conversation text, and the reorganized conversation is segmented into conversation segments with approximately n segments that can be processed by the model according to the length of the conversation and the processing effect of the model on the text. Then, the text classification model is used to determine whether it is an installation and maintenance related conversation. For those that are determined to be installation and maintenance related conversations, the segmented conversation segments are used as inputs of the reservation information extraction model, and the model output results are processed by the custom dictionary and filtering rules to obtain the minimum reservation information.

[0121] In one embodiment of the present disclosure, due to the long call time, the conversation text may include more characters, which may easily exceed the processing capacity of the traditional model, increasing the difficulty of the model to extract the appointment information; at the same time, since the ASR-translated call text often contains more colloquial content and typos, it will also seriously affect the extraction effect of the model. Figure 5 A flowchart of another method for extracting conversation text information in an embodiment of the present disclosure is shown as follows: Figure 5 As shown, the method comprises the following steps:

[0122] S501, obtaining a conversation text between a calling user and a called user.

[0123] In one embodiment of the present disclosure, the conversation text includes: a calling conversation text, a called conversation text, and a complete conversation text between the calling user and the called user.

[0124] S502, organize the acquired conversation contents into conversation pairs, and add a caller identifier caller: or a called identifier callee: to each corresponding conversation to mark whether each sentence belongs to the calling conversation text or the called conversation text, and use the reorganized conversation contents as the conversation text content union_text of the question-and-answer mode.

[0125] S503, input the caller conversation text into the pre-trained multi-classification model to determine whether the conversation text between the calling user and the called user is the conversation text of the reservation target service. If so, execute S504; if not, continue to execute S502 based on the latest acquired conversation text.

[0126] S504, calling a preset large language model prompt format, and generating a corresponding model prompt sentence according to union_text.

[0127] S505, input the model prompt sentence into a pre-trained large language model, and output summary information related to the target business.

[0128] In one embodiment of the present disclosure, the summary information needs to include necessary information elements related to installation and maintenance, which may include but are not limited to the installation and maintenance appointment visit time, appointment location, and the content of the installation and maintenance services required.

[0129] S506: Input summary information related to the target service into a pre-trained reservation information extraction model to obtain reservation information of the target service.

[0130] S507, matching the reservation information of the target service with the pre-built reservation information rule matching table to obtain the target reservation information.

[0131] As can be seen from the above, after obtaining the conversation text between the calling user and the called user, the embodiment of the disclosure determines whether the obtained conversation text is the conversation text of the reservation target service according to the conversation text of the calling user. If so, the complete conversation text between the calling user and the called user is marked according to the obtained conversation text to obtain multiple conversation text pairs, and then based on the pre-trained reservation information extraction model, the reservation information of the target service is extracted from the multiple conversation text pairs. The embodiment of the disclosure can improve the accuracy of obtaining reservation information from call recordings, thereby improving the service quality and customer satisfaction of installation and maintenance service personnel.

[0132] In one embodiment of the present disclosure, before executing the above S504, data cleaning may be performed on union_text to remove some stop words that do not affect the semantics, so that the extracted reservation information is more accurate.

[0133] In one embodiment of the present disclosure, the summary information is converted by word embedding technology to obtain a series of word vectors to form a word vector sequence. If there are n words in a sentence, n word vectors will be obtained, which can be expressed as X = {x1, x2, ..., x n}.

[0134] In one embodiment of the present disclosure, the model is mainly composed of an encoder, a decoder, and an extended vocabulary. The encoder uses a Bi-RNN network composed of a bidirectional gated recurrent unit Bi-GRU, and the decoder is composed of a GRU unidirectional RNN network with attention. The input sequence can be expressed as X = {x1, x2, ..., x n}, the output sequence can be expressed as Y = {y1,y2,…,y n}.

[0135] In one embodiment of the present disclosure, the encoder part involves the following calculation, and the state of the hidden layer of the encoder corresponding to the input sequence can be expressed as H = {h1, h2, ..., h n}:

[0136]

[0137] in, represents the forward hidden state at time t, represents the backward hidden state at time t, represents the forward GRU unit, represents the backward GRU unit, x t represents the input at time t of the input sequence, represents the forward hidden state at the previous moment, represents the backward hidden state at the previous moment, Indicates splicing, h t represents the final hidden state at time t.

[0138] In one embodiment of the present disclosure, the calculation involved in the decoder part can be calculated by the above formula (7), and the state of the decoder hidden layer can be expressed as D = {d1, d2, ..., d n}.

[0139] In one embodiment of the present disclosure, the attention score can be calculated by the above formula (1), the normalized attention weight can be calculated by the above formula (2), and the context vector can be calculated by the above formula (3).

[0140] In one embodiment of the present disclosure, a large language model is first used to convert the ASR-to-text conversation content, and summary information related to the target business is generated according to a preset large language model prompt format. If the conversation contains content related to installation and maintenance appointments for on-site services, the generated summary information needs to include installation and maintenance related information and appointment information.

[0141] In one embodiment of the present disclosure, by adopting a large language model to obtain summary information, the problem of installation and maintenance dialogue text being seriously affected by the spoken language of the dialogue, and the problem of typos in the text converted by ASR being affected by the conversion accuracy of the model are solved, and the phenomenon of the dialogue content being too long to be processed by the traditional model is reduced. At the same time, using the GRU model instead of the LSTM model reduces the parameters of the model and reduces the time cost of model training and prediction. The embodiment of the present disclosure can automatically and accurately extract installation and maintenance appointments, improve the service efficiency of installation and maintenance personnel, and thus improve customer satisfaction.

[0142] In one embodiment of the present disclosure, three conversation texts of call information between customers and installation and maintenance personnel collected by the voice data collection platform are first classified according to the content of the conversation to determine whether it is an installation and maintenance fault report. For the conversation content of installation and maintenance, the complete conversation is used as the input of the large language model. The large language model generates summary information containing installation and maintenance related information (which may include but is not limited to appointment time, fault, fault type, etc.) according to the pre-generated model prompt sentence. The generated summary information is used as the input of the appointment information extraction model, and the model output result is processed by the custom dictionary and filtering rules to obtain the minimum appointment information.

[0143] Based on the same inventive concept, the embodiment of the present disclosure also provides a device for extracting conversation text information, as described in the following embodiment. Since the principle of solving the problem in the device embodiment is similar to that in the above method embodiment, the implementation of the device embodiment can refer to the implementation of the above method embodiment, and the repeated parts will not be repeated.

[0144] Figure 6 A schematic diagram of a conversation text information extraction device in an embodiment of the present disclosure is shown. Figure 6 As shown, the device includes: a dialogue text acquisition module 601, a dialogue text judgment module 602, a dialogue text reconstruction module 603 and a reservation information extraction module 604.

[0145] Among them, the conversation text acquisition module 601 is used to obtain the conversation text between the calling user and the called user, wherein the conversation text includes: the calling conversation text, the called conversation text and the complete conversation text between the calling user and the called user; the conversation text judgment module 602 is used to determine whether the conversation text between the calling user and the called user is the conversation text for booking the target service according to the conversation text of the calling user; the conversation text reconstruction module 603 is used to mark the complete conversation text between the calling user and the called user according to the calling conversation text and the called conversation text to obtain multiple conversation text pairs, wherein each conversation text pair includes a calling conversation text and a called conversation text; the reservation information extraction module 604 is used to extract the reservation information of the target service from the multiple conversation text pairs based on the pre-trained reservation information extraction model.

[0146] As can be seen from the above, after obtaining the conversation text between the calling user and the called user, the embodiment of the disclosure determines whether the obtained conversation text is the conversation text of the reservation target service according to the conversation text of the calling user. If so, the complete conversation text between the calling user and the called user is marked according to the obtained conversation text to obtain multiple conversation text pairs, and then based on the pre-trained reservation information extraction model, the reservation information of the target service is extracted from the multiple conversation text pairs. The embodiment of the disclosure can improve the accuracy of obtaining reservation information from call recordings, thereby improving the service quality and customer satisfaction of installation and maintenance service personnel.

[0147] In one embodiment of the present disclosure, the above-mentioned reservation information extraction module 604 is also used to segment the multiple dialogue text pairs according to preset lengths to obtain multiple groups of dialogue text pairs, wherein each group of dialogue text pairs includes at least one dialogue text pair; each group of dialogue text pairs is sequentially input into a pre-trained reservation information extraction model until the reservation information extraction model outputs the reservation information of the target business, and the input of the segmented dialogue text pairs into the reservation information extraction model is stopped.

[0148] In one embodiment of the present disclosure, the device also includes: a summary information extraction module 605, which is used to obtain a preset prompt word format corresponding to a pre-trained large language model; call the preset prompt word format to generate a prompt sentence related to the target business according to multiple dialogue text pairs; input the prompt sentence related to the target business into the pre-trained large language model to obtain summary information related to the target business.

[0149] In one embodiment of the present disclosure, the above-mentioned appointment information extraction module 604 is further used to input the summary information related to the target service into a pre-trained appointment information extraction model to obtain the appointment information of the target service.

[0150] In one embodiment of the present disclosure, the device also includes: an appointment information preprocessing module 606, which is used to match the appointment information of the target service with a pre-constructed appointment information rule matching table to obtain the target appointment information, wherein the appointment information rule matching table is used to remove text in the appointment information of the target service that is not related to the target appointment information.

[0151] In one embodiment of the present disclosure, the target appointment information includes an appointment time.

[0152] Those skilled in the art will appreciate that various aspects of the present disclosure may be implemented as systems, methods or program products. Therefore, various aspects of the present disclosure may be specifically implemented in the following forms, namely: complete hardware implementation, complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software, which may be collectively referred to herein as "circuits", "modules" or "systems".

[0153] Refer to the following Figure 7 An electronic device 700 according to this embodiment of the present disclosure is described. Figure 7 The electronic device 700 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0154] like Figure 7 As shown, the electronic device 700 is in the form of a general computing device. The components of the electronic device 700 may include but are not limited to: at least one processing unit 710, at least one storage unit 720, and a bus 730 connecting different system components (including the storage unit 720 and the processing unit 710).

[0155] The storage unit stores a program code, and the program code can be executed by the processing unit 710, so that the processing unit 710 executes the steps described in the above "Exemplary Method" section of this specification according to various exemplary embodiments of the present disclosure. For example, the processing unit 710 can execute the following steps of the above method embodiment: obtaining a conversation text between a calling user and a called user, wherein the conversation text includes: a calling conversation text, a called conversation text, and a complete conversation text between the calling user and the called user; determining whether the conversation text between the calling user and the called user is a conversation text for booking a target service according to the conversation text of the calling user; if so, marking the complete conversation text between the calling user and the called user according to the calling conversation text and the called conversation text to obtain multiple conversation text pairs, wherein each conversation text pair includes a calling conversation text and a called conversation text; extracting the reservation information of the target service from the multiple conversation text pairs based on a pre-trained reservation information extraction model.

[0156] The storage unit 720 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 7201 and / or a cache storage unit 7202 , and may further include a read-only storage unit (ROM) 7203 .

[0157] The storage unit 720 may also include a program / utility 7204 having a set (at least one) of program modules 7205, such program modules 7205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0158] Bus 730 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0159] The electronic device 700 may also communicate with one or more external devices 740 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 700, and / or communicate with any device that enables the electronic device 700 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 750. Furthermore, the electronic device 700 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 760. As shown, the network adapter 760 communicates with other modules of the electronic device 700 via a bus 730. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 700, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, etc.

[0160] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0161] Based on the same inventive concept, the embodiment of the present disclosure also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, any of the above methods for extracting conversation text information is implemented. Since the principle of solving the problem in the embodiment of the computer-readable storage medium is similar to that in the above method embodiment, the implementation of the embodiment of the computer-readable storage medium can refer to the implementation of the above method embodiment, and the repeated parts will not be repeated.

[0162] More specific examples of computer-readable storage media in the present disclosure may include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0163] In the present disclosure, a computer readable storage medium may include a data signal propagated in baseband or as part of a carrier wave, wherein a readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A readable signal medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0164] Alternatively, the program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.

[0165] In a specific implementation, the program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., using an Internet service provider to connect through the Internet).

[0166] Based on the same inventive concept, a computer program product is also provided in the embodiments of the present disclosure, including a computer program product, including: a computer program or an instruction, wherein when the computer program or the instruction is executed by a processor, the conversation text information extraction method of any one of the above method embodiments is implemented. Since the principle of solving the problem in the computer program product embodiment is similar to that in the above method embodiment, the implementation of the computer program product embodiment can refer to the implementation of the above method embodiment, and the repeated parts will not be repeated.

[0167] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0168] In addition, although the steps of the method in the present disclosure are described in a specific order in the drawings, this does not require or imply that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps, etc.

[0169] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a mobile terminal, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0170] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to cover any variations, uses or adaptations of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The description and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the appended claims.

Claims

1. A method for extracting conversation text information, characterized in that: include: Acquire a conversation text between a calling user and a called user, wherein the conversation text includes: a calling conversation text, a called conversation text, and a complete conversation text between the calling user and the called user; Determining whether the conversation text between the calling user and the called user is a conversation text for a reserved target service according to the conversation text of the calling user; If yes, then marking the complete conversation text between the calling user and the called user according to the calling conversation text and the called conversation text to obtain a plurality of conversation text pairs, wherein each conversation text pair includes a calling conversation text and a called conversation text; Based on a pre-trained reservation information extraction model, the reservation information of the target service is extracted from the multiple dialogue text pairs.

2. The method for extracting conversation text information according to claim 1, characterized in that: The step of extracting the reservation information of the target service from the plurality of dialogue text pairs based on the pre-trained reservation information extraction model includes: Segmenting the plurality of dialogue text pairs according to a preset length to obtain a plurality of groups of dialogue text pairs, wherein each group of dialogue text pairs includes at least one dialogue text pair; Input each group of dialogue text pairs into the pre-trained reservation information extraction model in sequence until the reservation information extraction model outputs the reservation information of the target service, and stop inputting the segmented dialogue text pairs into the reservation information extraction model.

3. The method for extracting conversation text information according to claim 1, characterized in that: Before extracting the reservation information of the target service from the plurality of dialogue text pairs based on the pre-trained reservation information extraction model, the method further includes: Get the preset prompt word format corresponding to the pre-trained large language model; Calling the preset prompt word format to generate a prompt sentence related to the target business according to the multiple dialogue text pairs; The prompt sentence related to the target business is input into the pre-trained large language model to obtain summary information related to the target business.

4. The method for extracting conversation text information according to claim 3, characterized in that: The step of extracting the reservation information of the target service from the plurality of dialogue text pairs based on the pre-trained reservation information extraction model includes: The summary information related to the target business is input into a pre-trained reservation information extraction model to obtain the reservation information of the target business.

5. The method for extracting conversation text information according to claim 1, characterized in that: After extracting the reservation information of the target service from the plurality of dialogue text pairs based on the pre-trained reservation information extraction model, the method further includes: The reservation information of the target service is matched with a pre-built reservation information rule matching table to obtain the target reservation information, wherein the reservation information rule matching table is used to remove text irrelevant to the target reservation information from the reservation information of the target service.

6. The method for extracting conversation text information according to claim 5, characterized in that: The target reservation information includes a reservation time.

7. A device for extracting conversation text information, characterized in that: include: A conversation text acquisition module, used to acquire the conversation text between the calling user and the called user, wherein the conversation text includes: the calling conversation text, the called conversation text and the complete conversation text between the calling user and the called user; A conversation text determination module, used to determine whether the conversation text between the calling user and the called user is a conversation text for a reserved target service according to the conversation text of the calling user; a dialogue text reconstruction module, for marking the complete dialogue text between the calling user and the called user according to the calling dialogue text and the called dialogue text to obtain a plurality of dialogue text pairs, wherein each dialogue text pair includes a calling dialogue text and a called dialogue text; The reservation information extraction module is used to extract the reservation information of the target service from the multiple dialogue text pairs based on a pre-trained reservation information extraction model.

8. An electronic device, characterized in that: include: processor; as well as A memory, configured to store executable instructions of the processor; The processor is configured to execute the conversation text information extraction method according to any one of claims 1 to 6 by executing the executable instructions.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for extracting conversation text information described in any one of claims 1 to 6 is implemented.

10. A computer program product comprising: A computer program or instruction, characterized in that when the computer program or instruction is executed by a processor, it implements the method for extracting conversation text information as described in any one of claims 1 to 6.

Images