Service data processing method and device based on deep learning model

Abstract

The invention discloses a service data processing method and device based on a deep learning model. Relates to the technical field of artificial intelligence. The method comprises the following steps: receiving service request data input by a target object, and identifying a service type of the service request data to obtain a target service type; under the condition that the target service type is a consultation type, generating first reply data based on the target knowledge base and the service request data: under the condition that the target service type is a query type or a predetermined type, inputting the service request data into a target deep learning model to obtain intention data, and generating second reply data based on the intention data and the service request data, the second reply data including one of the following: a query result obtained by performing data query on the specified service product, and a predetermined result obtained by performing predetermined operation on the specified service product. According to the invention, the technical problem of poor processing effect of the request data of the additional service in the ticket booking process in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a service data processing method and device based on a deep learning model. Background Art

[0002] At present, how to provide more convenient and intelligent additional service product booking and inquiry for passengers' reference and selection has become the key to improving airline services from good to excellent. In the relevant technology, online booking and inquiry of additional services have the following shortcomings:

[0003] First, when inquiries and reservations are made by filling out forms, users need to manually select and fill out various information. This operation is cumbersome, error-prone, and inconvenient, resulting in low efficiency in processing user service requests and reducing travelers' willingness to buy.

[0004] Second, it cannot automatically respond to users' business questions. It is only equipped with a fault and problem solving ticket system and a dedicated manual service desk. Problems encountered during the query and booking of additional services cannot be answered in a timely manner. Moreover, the ticket system needs to go through layers of circulation, and the response is not timely. The manual service desk is limited by human resources and cannot promptly solve the problems encountered by users when booking tickets and inquiries. Therefore, it cannot promptly answer users' questions, resulting in low efficiency in processing user service requests.

[0005] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0006] The embodiments of the present invention provide a service data processing method and device based on a deep learning model, so as to at least solve the technical problem of poor processing effect of request data for additional services during the ticket booking process in the related technology.

[0007] According to one aspect of an embodiment of the present invention, a service data processing method based on a deep learning model is provided, comprising: receiving service request data input by a target object, and identifying the service type of the service request data to obtain a target service type, wherein the service type of the service request data includes one of the following: a consulting class for consulting a service product, a query class for querying data of the service product, and a reservation class for reserving the service product; when the target service type is the consulting class, generating first reply data based on a target knowledge base and the service request data, wherein the target knowledge base includes at least: M questions associated with the service product and the reply content of each question, and the first reply data The complex data is used to reply to the questions carried in the service request data, and M is a positive integer: when the target service type is the query class or the reservation class, the service request data is input into the target deep learning model to obtain intent data, and second reply data is generated based on the intent data and the service request data, wherein the target deep learning model is used to identify the intention of the target object, and the intention of the target object includes one of the following: the need to perform data query on the specified service product, and the need to reserve the specified service product, and the second reply data includes one of the following: the query result of the data query on the specified service product, and the reservation result of the reservation of the specified service product.

[0008] Furthermore, based on the target knowledge base and the service request data, first reply data is generated, including: based on the word frequency-inverse document frequency algorithm, determining the similarity between the questions carried in the service request data and each question in the target knowledge base to obtain a similarity set; based on the similarity set, extracting N target questions in the target knowledge base, wherein the similarity between the target question and the question carried in the service request data is greater than the similarity between the questions in the target knowledge base other than the target question and the question carried in the service request data, and N is a positive integer less than M; based on the N target questions, the first reply data is generated.

[0009] Furthermore, based on the N target questions, the first reply data is generated, including: inputting the N target questions and the service request data into a first deep learning model, outputting a score value for each target question, and determining the highest score value of the N target questions, wherein the first deep learning model is used to identify the score value of the similarity between the target question and the question carried in the service request data, and the score value of each target question is used to represent the size of the similarity between the target question and the question carried in the service request data; when the highest score value of the N target questions is greater than a preset threshold, the first reply data is generated based on the reply content corresponding to the target question associated with the highest score value.

[0010] Furthermore, based on the word frequency-inverse document frequency algorithm, the similarity between the questions carried in the service request data and each question in the target knowledge base is determined to obtain a similarity set, including: performing word segmentation processing on the questions carried in the service request data to obtain a word segmentation result; based on the word frequency-inverse document frequency algorithm, performing vectorization processing on the word segmentation result to obtain a vectorized word segmentation result; calculating the cosine similarity between the vectorized word segmentation result and each question in the target knowledge base to obtain a similarity set.

[0011] Further, generating second reply data based on the intent data and the service request data includes: determining a preset slot based on the intent data, wherein the preset slot includes: attributes of a field required to perform a target operation, and the target operation includes one of the following: performing a data query on the specified service product, and reserving the specified service product; filling a slot value of the preset slot based on the service request data to obtain a first slot filling result, wherein the slot value of the preset slot represents an attribute value of the attribute of the field required to perform the target operation; executing the target operation based on the first slot filling result to obtain an execution result, and generating the second reply data based on the execution result, wherein the data format of the execution result is different from the data format of the second reply data.

[0012] Further, based on the first slot filling result, the target operation is executed, including: detecting whether there is a missing slot value in the preset slot to obtain a detection result; when the detection result indicates that there is no missing slot value in the preset slot, executing the target operation based on the first slot filling result; when the detection result indicates that there is a missing slot value in the preset slot, obtaining the missing slot value in the preset slot by conducting multiple rounds of dialogues with the target object to obtain the target slot value, and filling the missing slot value in the preset slot based on the target slot value to obtain a second slot filling result; executing the target operation based on the second slot filling result.

[0013] Further, identifying the service type of the service request data to obtain a target service type includes: identifying the service type of the service request data based on preset rules to obtain an identification result; when the identification result indicates that the service type of the service request data has been identified unsuccessfully, inputting the service request data into a second deep learning model to output the target service type, wherein the second deep learning model is used to identify the service type of the service request data; and when the identification result indicates that the service type of the service request data has been identified successfully, determining the target service type based on the identification result.

[0014] Furthermore, after generating the first reply data based on the target knowledge base, or after generating the second reply data based on the intention of the target object, it also includes: acquiring the evaluation data of the target object, and determining the satisfaction of the target object based on the evaluation data.

[0015] According to another aspect of an embodiment of the present invention, a service data processing device based on a deep learning model is also provided, including: a first processing unit, used to receive service request data input by a target object, and identify the service type of the service request data to obtain a target service type, wherein the service type of the service request data includes one of the following: a consultation class for consulting a service product, a query class for querying data of the service product, and a reservation class for reserving the service product; a generation unit, used to generate first reply data based on a target knowledge base and the service request data when the target service type is the consultation class, wherein the target knowledge base at least includes: M questions associated with the service product and the reply content of each question, The first reply data is used to reply to the question carried in the service request data, and M is a positive integer: the second processing unit is used to input the service request data into the target deep learning model when the target service type is the query class or the reservation class, to obtain intent data, and generate second reply data based on the intent data and the service request data, wherein the target deep learning model is used to identify the intention of the target object, and the intention of the target object includes one of the following: the need to perform data query on the specified service product, and the need to reserve the specified service product, and the second reply data includes one of the following: the query result of the data query on the specified service product, and the reservation result of the reservation of the specified service product.

[0016] Further, the generation unit includes: a first determination subunit, used to determine the similarity between the questions carried in the service request data and each question in the target knowledge base based on a word frequency-inverse document frequency algorithm to obtain a similarity set; an extraction subunit, used to extract N target questions in the target knowledge base based on the similarity set, wherein the similarity between the target question and the question carried in the service request data is greater than the similarity between the questions in the target knowledge base other than the target question and the question carried in the service request data, and N is a positive integer less than M; a generation subunit, used to generate the first reply data based on the N target questions.

[0017] Furthermore, the generation subunit includes: a first processing module, used to input N of the target questions and the service request data into a first deep learning model, output a score value for each of the target questions, and determine the highest score value of the N target questions, wherein the first deep learning model is used to identify the score value of the similarity between the target question and the question carried in the service request data, and the score value of each of the target questions is used to indicate the size of the similarity between the target question and the question carried in the service request data; a generation module, used to generate the first reply data based on the reply content corresponding to the target question associated with the highest score value when the highest score value of the N target questions is greater than a preset threshold.

[0018] Furthermore, the first determination subunit includes: a second processing module, used to perform word segmentation processing on the questions carried in the service request data to obtain a word segmentation result; a third processing module, used to vectorize the word segmentation result based on the word frequency-inverse document frequency algorithm to obtain a vectorized word segmentation result; a calculation module, used to calculate the cosine similarity between the vectorized word segmentation result and each question in the target knowledge base to obtain a similarity set.

[0019] Further, the second processing unit includes: a second determination sub-unit, used to determine a preset slot based on the intention data, wherein the preset slot includes: attributes of a field required to perform a target operation, and the target operation includes one of the following: data query for the specified service product, reservation for the specified service product; a filling sub-unit, used to fill the slot value of the preset slot based on the service request data to obtain a first slot filling result, wherein the slot value of the preset slot represents an attribute value of the attribute of the field required to perform the target operation; a first processing sub-unit, used to execute the target operation based on the first slot filling result, obtain an execution result, and generate the second reply data based on the execution result, wherein the data format of the execution result is different from the data format of the second reply data.

[0020] Further, the processing sub-unit includes: a detection module, used to detect whether there is a missing slot value in the preset slot and obtain a detection result; a first execution module, used to perform the target operation based on the first slot filling result when the detection result indicates that there is no missing slot value in the preset slot; a fourth processing module, used to obtain the missing slot value in the preset slot by conducting multiple rounds of dialogues with the target object when the detection result indicates that there is a missing slot value in the preset slot, obtain the target slot value, and fill the missing slot value in the preset slot based on the target slot value to obtain a second slot filling result; a second execution module, used to execute the target operation based on the second slot filling result.

[0021] Further, the first processing unit includes: an identification subunit, used to identify the service type of the service request data based on preset rules to obtain an identification result; a second processing subunit, used to input the service request data into a second deep learning model and output the target service type when the identification result indicates that the service type identification of the service request data has failed, wherein the second deep learning model is used to identify the service type of the service request data; and a third determination subunit, used to determine the target service type based on the identification result when the identification result indicates that the service type identification of the service request data has been successful.

[0022] Furthermore, the service data processing device based on the deep learning model also includes: a third processing unit, which is used to obtain evaluation data of the target object after generating first reply data based on the target knowledge base, or after generating second reply data based on the intention of the target object, and determine the satisfaction of the target object based on the evaluation data.

[0023] According to another aspect of an embodiment of the present invention, an electronic device is also provided, including: 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 service data processing methods based on a deep learning model by executing the executable instructions.

[0024] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is provided, which stores a computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the above-mentioned service data processing methods based on deep learning models.

[0025] In the present invention, service request data input by a target object is received, and a service type of the service request data is identified to obtain a target service type, wherein the service type of the service request data includes one of the following: a consultation class for consulting a service product, a query class for querying data of a service product, and a reservation class for reserving a service product; when the target service type is a consultation class, first reply data is generated based on a target knowledge base and the service request data, wherein the target knowledge base includes at least: M questions associated with the service product and the reply content of each question, and the first reply data is used to reply to the questions carried in the service request data, where M is a positive integer: when the target service type is a query class or a reservation class, the service request data is input into a target deep learning model to obtain intent data, and second reply data is generated based on the intent data and the service request data, wherein the target deep learning model is used to identify the intent of the target object, and the intent of the target object includes one of the following: the need to perform a data query on a specified service product, the need to reserve a specified service product, and the second reply data includes one of the following: a query result of performing a data query on a specified service product, a reservation result of reserving a specified service product. This solves the technical problem of poor processing effect of request data for additional services in the ticket booking process in the related technology.

[0026] In the present invention, after receiving the service request data, the service request data of the additional service is automatically processed based on the service type and the deep learning model and the corresponding reply data is generated, thereby avoiding the low processing efficiency of processing the service request data of the additional service by filling in forms or manual service in the related technology, thereby achieving the technical effect of improving the processing efficiency of the service request data of the additional service. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0028] Figure 1 is a flowchart of an optional service data processing method based on a deep learning model according to an embodiment of the present invention;

[0029] Figure 2 is a flowchart of an optional service data processing process based on a deep learning model according to an embodiment of the present invention;

[0030] Figure 3 is a schematic diagram of an optional service data processing system based on a deep learning model according to an embodiment of the present invention;

[0031] Figure 4is a schematic diagram of an optional service data processing device based on a deep learning model according to an embodiment of the present invention;

[0032] Figure 5 is a schematic diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0033] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0034] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0035] It should be noted that the user information (including but not limited to user device information, user personal information, etc.), collected information and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with the relevant laws, regulations and standards of the relevant regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0036] Embodiment 1

[0037] According to an embodiment of the present invention, an optional method embodiment of a service data processing method based on a deep learning model is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in an order different from that shown here.

[0038] Figure 1 is a flowchart of an optional service data processing method based on a deep learning model according to an embodiment of the present invention. Figure 1 As shown, the method comprises the following steps:

[0039] Step S101, receiving service request data input by the target object, and identifying the service type of the service request data to obtain the target service type, wherein the service type of the service request data includes one of the following: a consultation class for consulting service products, a query class for querying data of service products, and a reservation class for reserving service products.

[0040] The types of the above-mentioned service products may be: additional services of transportation, additional services provided to passengers by airlines, such as seat selection, meals, luggage, etc. The above-mentioned target objects may be objects that consult, inquire or book the specified service products, such as passengers who consult, inquire or book additional services of the target transportation. The above-mentioned service request data may include: a request for consulting, inquiring or booking the specified service products. In this embodiment, the service request data may be input into a specified dialog box, and the service request data may be input by means of a dialogue through the dialog box.

[0041] The above-mentioned consultation category may refer to a service type for consulting on the concept of a service product, the query category may refer to a service type for querying data of a service product, and the reservation category may refer to a service type for reserving a service product.

[0042] In this embodiment, the service type of the service request data may be first identified according to preset rules. If the identification fails, the service type of the service request data may be classified by a pre-trained deep learning model.

[0043] Step S102, when the target service type is consulting, generates first reply data based on the target knowledge base and the service request data, wherein the target knowledge base includes at least: M questions associated with the service product and the reply content of each question, and the first reply data is used to reply to the questions carried in the service request data, and M is a positive integer.

[0044] The above-mentioned target knowledge base may include M pre-recorded questions associated with the service product and the response content of each question. In this embodiment, the first response data can be determined based on the question in the target knowledge base that has the highest similarity to the consulting question carried in the service request data. For example, if the highest similarity between the consulting question in the target knowledge base and the consulting question carried in the service request data is greater than a preset threshold, the response content corresponding to the question with the highest similarity can be used as the first response data.

[0045] Step S103, when the target service type is a query type or a reservation type, the service request data is input into the target deep learning model to obtain intent data, and second reply data is generated based on the intent data and the service request data, wherein the target deep learning model is used to identify the intention of the target object, and the intention of the target object includes one of the following: the need to perform a data query on a specified service product, and the need to make a reservation for a specified service product, and the second reply data includes one of the following: the query result of the data query on the specified service product, and the reservation result of the reservation of the specified service product.

[0046] The above-mentioned intention data may be a data query for a specified service product that the target object needs to query, or the above-mentioned intention data may be a reservation for a specified service product that the target object needs to make.

[0047] When the target service type is a query or reservation type, the service request data can be input into the target deep learning model to obtain the intent data.

[0048] In the case where the intention data indicates that the target object needs to query the data of a specified service product, the data of the specified service product is queried, and then second reply data can be generated based on the query result.

[0049] In the case where the intention data indicates that the target object needs to reserve a specified service product, a reservation can be made for the specified service product, and then second reply data can be generated based on the reservation result.

[0050] In this embodiment, through the above steps, after receiving the service request data, the service request data of the additional service is automatically processed based on the service type and the deep learning model and the corresponding reply data is generated, avoiding the low processing efficiency of the service request data of the additional service by filling in a form or manual service in the related art, thereby achieving the technical effect of improving the processing efficiency of the service request data of the additional service. In addition, the technical problem of poor processing effect of the request data of the additional service in the ticket booking process in the related art is solved.

[0051] Optionally, based on the target knowledge base and the service request data, first reply data is generated, including: based on the word frequency-inverse document frequency algorithm, determining the similarity between the question carried in the service request data and each question in the target knowledge base to obtain a similarity set; based on the similarity set, extracting N target questions in the target knowledge base, wherein the similarity between the target question and the question carried in the service request data is greater than the similarity between questions other than the target question in the target knowledge base and the question carried in the service request data, and N is a positive integer less than M; based on the N target questions, the first reply data is generated.

[0052] For example, if the service request data input by the target object belongs to the business consulting category, the TFIDF (term frequency-inverse document frequency algorithm) algorithm can be used to perform a recall operation in the business knowledge base to retrieve the N questions with the highest similarity to the user questions, and then input them into a pre-trained deep learning model (for example, the Bert model (a pre-trained language model)) to find the question with the highest similarity score. If the score of the question with the highest similarity score is lower than a preset threshold (for example, 0.5), the first reply data can be that the user cannot find the question and can transfer it to manual. Otherwise, the first reply data can be the reply content of the question with the highest similarity score, thereby achieving the technical effect of improving the reply efficiency of additional service request data of the consulting category.

[0053] Optionally, based on N target questions, first reply data is generated, including: inputting the N target questions and service request data into a first deep learning model, outputting a score value for each target question, and determining the highest score value of the N target questions, wherein the first deep learning model is used to identify the score value of the similarity between the target question and the question carried in the service request data, and the score value of each target question is used to indicate the size of the similarity between the target question and the question carried in the service request data; when the highest score value of the N target questions is greater than a preset threshold, the first reply data is generated based on the reply content corresponding to the target question associated with the highest score value.

[0054] The above-mentioned first deep learning model can be used to identify the score value of the similarity between the target question and the question carried in the service request data. The model type of the first deep learning model can be a Bert model. The first deep learning model may include: using multiple predefined questions and similar questions to each question as training samples, and the similarity between the predefined questions and the similar questions to each question as training labels to train the Bert model that has not been trained, to obtain a deep learning model.

[0055] In this embodiment, after N target questions are screened out based on the word frequency-inverse document frequency algorithm, the N target questions and service request data can be input into the first deep learning model to output the precise similarity score value between each target question and the question carried in the service request data. Then, when the highest score value of the N target questions is greater than a preset threshold, the reply content corresponding to the target question associated with the highest score value can be used as the first reply data. When the highest score value of the N target questions is not greater than the preset threshold, the first reply data can be transferred to manual if the user cannot query the question, thereby achieving the purpose of accurately replying to the service request data of the target object.

[0056] Optionally, based on the word frequency-inverse document frequency algorithm, the similarity between the questions carried in the service request data and each question in the target knowledge base is determined to obtain a similarity set, including: performing word segmentation on the questions carried in the service request data to obtain a word segmentation result; based on the word frequency-inverse document frequency algorithm, vectorizing the word segmentation result to obtain a vectorized word segmentation result; calculating the cosine similarity between the vectorized word segmentation result and each question in the target knowledge base to obtain a similarity set.

[0057] For example, if the service request data input by the target object belongs to the business consulting category, the TFIDF (term frequency-inverse document frequency algorithm) algorithm can be used in the business knowledge base to perform a recall operation to retrieve the N questions with the highest similarity to the user questions. Specifically, the questions carried in the service request data can be segmented to obtain segmentation results, and then the segmentation results can be vectorized using the term frequency-inverse document frequency algorithm to obtain vectorized segmentation results. The cosine similarity between the vectorized segmentation results and each question in the target knowledge base can be calculated. For example, the cosine similarity between the vectorized segmentation results and the vectorized results of each question in the target knowledge base can be calculated to obtain a similarity set, thereby achieving the purpose of accurately calculating the similarity between the consulting questions carried in the service request and the questions in the target knowledge base.

[0058] Optionally, generating second reply data based on the intent data and the service request data includes: determining a preset slot based on the intent data, wherein the preset slot includes: attributes of a field required to perform a target operation, the target operation including one of the following: performing a data query on a specified service product, reserving a specified service product; filling a slot value of the preset slot based on the service request data to obtain a first slot filling result, wherein the slot value of the preset slot represents an attribute value of the attribute of the field required to perform the target operation; executing the target operation based on the first slot filling result to obtain an execution result, and generating second reply data based on the execution result, wherein the data format of the execution result is different from the data format of the second reply data.

[0059] The above-mentioned preset slots may include: attributes of the fields required to perform the target operation. For example, the preset slots may include but are not limited to query services, departure point, arrival point, departure time, cabin class, seats and other attributes. The slot value may be a specific value of the slot, for example, the specific location of the departure point, the specific location of the arrival point, the specific time of the departure time and other attribute values. The data format of the above-mentioned execution results may be in XML (Extensible Markup Language) format, and the second reply data may be in natural language, including but not limited to: Chinese, English, etc.

[0060] After the intention of the target object is clear, a serialized request is assembled according to the specific intent data and intent slot values, and the query and reservation operations of the additional services (corresponding to the target operation) are executed. After the serialized results (corresponding to the execution results) returned by the query or reservation system, they can be translated into natural language (i.e., the second reply data is obtained) according to preset rules and returned.

[0061] For example, the slot value of the preset slot can be filled based on the service request data to obtain a first slot filling result, and then the corresponding additional service reservation or query system request can be generated based on the identified user intention and the filled slot value, and sent to the additional service computing system or the additional service reservation system for processing. The additional service reservation system or the additional service computing system performs corresponding query and reservation operations and returns the result (corresponding to the execution result). The returned result can generate natural language based on a rule method (that is, obtain the second reply data), wherein the returned result can be serialized XML format data.

[0062] Optionally, based on the first slot filling result, the target operation is performed, including: detecting whether there is a missing slot value in the preset slot to obtain a detection result; when the detection result indicates that there is no missing slot value in the preset slot, performing the target operation based on the first slot filling result; when the detection result indicates that there is a missing slot value in the preset slot, obtaining the missing slot value in the preset slot by conducting multiple rounds of dialogue with the target object to obtain the target slot value, and filling the missing slot value in the preset slot based on the target slot value to obtain a second slot filling result; performing the target operation based on the second slot filling result.

[0063] Requests for querying and booking additional services require the identification and filling of intended slots, and the specific fields (corresponding to preset slots) and field values ​​(i.e. slot values) required for querying and booking additional services are clarified based on the user's input. The additional service reservation system and the additional service query system can preset different slots in the system. For example, the slots required for querying additional services may include but are not limited to query service type, departure place, destination, departure date, cabin, frequent flyer level, etc.

[0064] In this embodiment, a method based on the deep learning model Bert can be used for processing. After the request is input into the Bert model, named entity recognition is performed to identify the slot and slot value. The format of the slot recognition output is:

[0065]

[0066] For example, the identified entity is the departure point Beijing, the arrival point Shanghai, and the query service is luggage. The intent and slot after identification are:

[0067]

[0068]

[0069] After the dialogue management module (used to communicate with the target object) collects information, it finds that the departure time slot is missing. Therefore, it can actively initiate the next dialogue with the target object to clarify the departure time. The dialogue management module automatically asks the user "When do you want to leave?" based on the missing departure time slot.

[0070] After receiving the message, the target object replies "January 8, 2024", and the time slot information is clear. If other required information is missing, you can continue to perform multiple conversations until the conversation management module has clarified all the required slot values ​​for querying additional services. For example, the slots and slot values ​​identified this time are finally as follows:

[0071]

[0072] Since it is impossible for a deep learning model to learn all slot recognition at once, slot recognition information may be lost during the recognition process. At this time, the dialogue management module can prompt the target to switch to manual processing, record the failure, and wait for the operation module to collect data to facilitate the subsequent improvement of slot filling accuracy.

[0073] Optionally, identifying the service type of the service request data to obtain a target service type includes: identifying the service type of the service request data based on preset rules to obtain an identification result; when the identification result indicates that identification of the service type of the service request data has failed, inputting the service request data into a second deep learning model and outputting the target service type, wherein the second deep learning model is used to identify the service type of the service request data; when the identification result indicates that identification of the service type of the service request data has been successful, determining the target service type based on the identification result.

[0074] In this embodiment, the target object can use a personal account to log in to the sales front end of the additional service, and then enter the additional service intelligent query and reservation entrance, and enter the operation text (corresponding to the service request data) to be performed, including querying available additional services, booking available additional services, business consultation, etc., and then the task classification (including business consultation areas or reservations, inquiries) input by the target object can be analyzed according to the natural language input by the target object (corresponding to the service request data). First, the classification is performed through the preset business rules (corresponding to the preset rules). If the preset business rules can be matched, the service type of the service request data can be directly obtained according to the business rules.

[0075] Due to the diversity of natural languages, the preset rules may not be matched completely. If not, the second deep learning model can be used to perform classification operations and analyze the classification of this task. According to the task type (corresponding to the service type), the request is distributed to the business consulting module or the task processing module, wherein the model type of the pre-trained second deep learning model can be a Bert model, and the second deep learning model can include: using historical service request data of multiple historical time periods as training samples, and the service type of each historical service request data as a training label. After model training of the deep learning model to be trained, the model obtained achieves the technical effect of improving the accuracy of the classification results of the service request data and the classification efficiency.

[0076] Optionally, after generating the first reply data based on the target knowledge base, or after generating the second reply data based on the intention of the target object, it also includes: acquiring evaluation data of the target object, and determining the satisfaction of the target object based on the evaluation data.

[0077] In this embodiment, after the conversation with the target object ends (after the first reply data is generated or the second reply data is generated), information such as service requests that fail to automatically identify the service type and negative comments from the target object can be regularly collected and published to the backend operation module. After the operator annotates the data, it can be input into the specified Bert model (i.e., the second deep learning model), and the model can be fine-tuned regularly to achieve a deep learning model that better fits the identification requirements of the service type. The problem that the related art cannot perform after-sales feedback analysis on additional services queried and booked by users is solved. The problem that the processing logs of additional services in the related art are massive and it is difficult to manually process user satisfaction feedback and after-sales analysis is avoided, thereby achieving the purpose of effective after-sales analysis of additional services.

[0078] Figure 2 is a flowchart of an optional service data processing process based on a deep learning model according to an embodiment of the present invention. Figure 2 As shown, including:

[0079] Step 1: Start the smart dialogue. After the target object (e.g., user) logs in to the additional service sales front end using a personal account, enters the additional service smart query and booking portal, and enters the text of the operation to be performed, including querying available additional services, booking available additional services, and business consultation.

[0080] Step 2: Classify the task (query / booking / consultation) input by the user. The service data processing system based on the deep learning model analyzes the task classification (corresponding to the service type, including business consulting field or booking, query) input by the target object according to the natural language input by the target object. First, the classification is performed through the preset business rules. Due to the diversity of natural language, the preset rules cannot be matched completely. If they cannot be found, the Bert model is used to perform the classification operation and analyze the task classification. According to the task type, the request is distributed to the business consulting module or the task processing module.

[0081] Step 3: Multiple rounds of dialogue clarify the user's intention. If the target object input belongs to the target object's business consulting field, the TFIDF algorithm is used in the business knowledge base to perform a recall operation to extract the ten questions with the highest similarity to the target object, and then input them into the Bert model to find the answer with the highest score similarity. If the probability of the answer score is less than 0.5, the target object is replied that the question cannot be found and can be transferred to manual, otherwise the answer is returned to the dialogue management module. If the target object input needs to perform specific additional service reservations and inquiries, the dialogue is input into the Bert model for intent recognition. The content of recognition includes but is not limited to querying or booking services, the departure point, arrival point, departure time, cabin class, seat attributes, etc. of the service. If it is found that a certain intention has not been clarified, the dialogue can be continued to clarify the intention.

[0082] Step 4, reservation / query / knowledge query, after the target object’s intention is clear, it is assembled into a serialized request based on the specific intention and intention slot value and sent to the query or reservation module of the additional service.

[0083] Step 5: The result of the reservation / query / knowledge query operation is returned, and the additional service query or reservation system performs the query and reservation operation of the additional service. After the operation is completed, the natural language processing module is used to generate a natural language expression and return it to the dialogue management module.

[0084] Step 6: The returned result is translated into natural language. After receiving the serialized result returned by the query or reservation system, the natural language processing module needs to translate it into natural language and return it according to preset rules.

[0085] Step 7: Collect user experience feedback and operation feedback. After the conversation ends, the operation module will regularly collect information such as target object requests and negative comments that cannot be automatically classified and identified, and publish it to the backend operation module. After the operation staff labels the data, they input it into the Bert model and regularly perform model fine-tuning to make the model more suitable for the existing system requirements.

[0086] Through this embodiment, users no longer need to fill out forms, but can complete inquiries and reservations through dialogue and natural language. The user input can be automatically classified based on the deep learning model trained and fine-tuned by BERT to automatically classify user needs, and inquire or book additional services or business consultation according to the classification results. Only a small number of unrecognizable user needs can be redirected to the manual customer service system for resolution, which can respond to user needs in real time and solve the problem of insufficient manpower.

[0087] The additional service query system and the additional service reservation system can proactively initiate conversations to clarify the information required for reservation and query operations. Customers can ask questions at any time and get responses to any problems encountered during the reservation process. After the reservation or query is completed, the user can give feedback on the service evaluation. The operation module will regularly collect feedback, input the deep learning model to perform sentiment classification, classify and collect the negative reviews, and distribute them to the operation staff. After analysis and data labeling by the operation staff, the Bert model can be trained. It solves the problem that the reservation and query operations of additional services in related technologies are cumbersome and prone to errors, and cannot improve user satisfaction. It avoids the lack of intelligence in the overall additional service in related technologies, improves the availability of the system, can provide real-time help to users, solve the problem of insufficient manpower, and reduce costs.

[0088] Embodiment 2

[0089] Embodiment 2 of the present invention provides an optional service data processing system based on a deep learning model (hereinafter referred to as the system), which can be used to execute the service data processing method based on a deep learning model provided in Embodiment 1.

[0090] Figure 3 is a schematic diagram of an optional service data processing system based on a deep learning model according to an embodiment of the present invention. Figure 3 As shown, it includes: an additional service sales front end, an additional service query cluster, an additional service booking cluster, a dialogue management module, a natural language generation module and an operation module.

[0091] Additional service sales front-end: including but not limited to the Air China sales front-end, airport self-service counters, etc., initiating dialogues with users and being responsible for dialogue management.

[0092] Additional service query cluster: receives requests from the additional service sales front end and returns the calculated results to the additional service sales front end.

[0093] Additional service reservation cluster: receives reservation requests from the additional service sales front end, reserves additional services, and returns the reservation results to the caller.

[0094] Dialogue management module: receives the natural language input from the target object, including but not limited to using methods based on preset rules and using the Bert model classification, analyzing the task classification of this dialogue (corresponding to the service type), distributing the request to the business consultation processing module or the task processing module, and managing multiple rounds of dialogues to clarify and proactively ask for unclear information in the user's query and booking of additional services.

[0095] Task processing module: Receive natural language text input by the user, input it into the Bert deep learning model, perform named entity recognition, and identify the category, intent, and intent slot of this conversation. Categories include querying additional services, booking additional services, and consulting. Intents include but are not limited to querying services, departure points, arrival points, departure time, cabin class, seat attributes, etc. Intent slots are specific values ​​in the identified intent.

[0096] Business consultation processing module: receives natural language text input by users, searches for and matches similar questions in the business knowledge base, including but not limited to introductions to additional services, additional service inquiries, and professional field questions in additional service reservations, and returns the answer with the highest similarity to the question in the knowledge base.

[0097] Natural language generation module: including but not limited to translating the serialized results of the additional service reservation system and the additional service query system into natural language in a rule-based manner, and generating natural language according to the intent slots that need to be clarified.

[0098] Operation module: The operation module continuously collects system logs and sends the logs that have not been processed correctly to the operation personnel. The operation continuously annotates the user conversations and feedback (including but not limited to unidentified task classifications, negative user reviews, etc.) to upgrade and optimize the deep network learning model and build a richer business knowledge base.

[0099] The processes that a service data processing system based on a deep learning model can perform include:

[0100] Step 1. User input request: The user (corresponding to the target object) submits a query or additional service booking request (corresponding to service request data) to the system in the form of text input. For example, the user submits a request of "check how much free baggage can be taken on flights from Beijing to Shanghai?"

[0101] Step 2: Automatically classify user tasks (i.e., identify service types). After receiving user input, it is necessary to analyze whether the input belongs to consulting service content or to perform reservation or inquiry service operations.

[0102] Table 1 is an example of preset rules (Note: "*" represents one or more arbitrary characters).

[0103] Table 1

[0104] Serial number Label Preset Rules 1 consult What is Free Baggage*? 2 consult What are the seat options*? 3 consult What does *cabin* mean? 4 Book now Book * Luggage * Kg 5 Query How much luggage did you find? 6 Book now Buy * Luggage * Kg

[0105] For example, if the input request is "check how much free baggage I can bring on flights from Beijing to Shanghai?", the system's preset query business rule 5 "check *how much *baggage* I can bring" can be matched, so it can be determined that this request belongs to the query operation (corresponding to the query category). However, if there is a request "I brought luggage of about 158CM in size, how should I buy luggage service?", the rule cannot be adapted.

[0106] The unmatched request is input into the trained Bert model (corresponding to the second deep learning model) to perform classification operations (i.e., identify the service type). The model can output the probabilities of the three categories (query, reservation, and business consultation). Taking the probabilities of 0.7, 0.2, and 0.1 as examples, the probability of the query type is 0.7, which is the largest. Therefore, it can be determined that this request is an additional service query operation, and the following "Step 3 Additional Service Query and Reservation Intention Identification" operation is performed. If it is a business consultation, go to "Step 6. Additional Service Business Knowledge Base Query" operation.

[0107] Step 3. Intent identification for additional service enquiries and bookings:

[0108] Requests to query and book additional services require the identification and filling of intent slots.

[0109] Intent recognition is to determine the user's true intention based on the user's input content, including intent slot recognition and slot filling. For the intelligent system, the specific operation is to clarify the specific fields (corresponding to the preset slots) and field values ​​(corresponding to the slot values) required by the additional service query module or booking module based on the user's input. Different slots are preset in the system for booking and inquiry. For example, the slots required for inquiring additional services include but are not limited to query service type, departure place, destination, departure date, cabin, frequent flyer level, etc.

[0110] Intent recognition is processed based on the deep learning model Bert. After the request is input into the Bert model, named entity recognition is performed to identify the slot and slot value. The format of slot recognition output is:

[0111]

[0112] For example, the entities identified in "Step 1. User input request" are departure Beijing, arrival Shanghai, and query service baggage. The intent and slot after identification are:

[0113]

[0114]

[0115] After the dialogue management module collects the information, it finds that the departure time slot is missing. Therefore, it can actively initiate the next dialogue to clarify the departure time. The dialogue management module automatically asks the user "When do you want to depart?" based on the missing departure time slot.

[0116] After receiving the message, the user replies "January 8, 2024", and the time slot information is clear. If other required information is still missing, continue to perform multiple conversations until the conversation management module clarifies all the required slot values ​​for querying additional services. For example, the slots and slot values ​​identified this time are finally as follows:

[0117]

[0118] Due to the complexity and variability of natural language, it is impossible for a deep learning model to learn all slot recognition at once. Therefore, slot recognition information may be lost during the recognition process. At this time, the dialogue management system prompts the user to switch to manual processing, records the failure, and waits for the operation module to collect data.

[0119] Step 4. Generate reservation and query requests: Generate the corresponding request required by the additional service reservation system or additional service query system according to the identified user intention and the filled slot value, and send it to the additional service query system or additional service reservation system for processing.

[0120] Step 5. Additional service reservation or computing system performs reservation or query operation: The additional service reservation or computing system performs corresponding query and reservation operation based on the information in the request, and returns the result. The returned result needs to be processed by the language processing module based on the rule method, and the natural language is generated and returned to the dialogue management module. For example, the result returned by the system is a serialized XML result, such as the request entered in step 1. Then the serialized XML result can be converted into natural language, and "two pieces of luggage can be carried for free on the flight" is returned.

[0121] Step 6. Query the business knowledge base for additional services: If the user input belongs to the user's business consultation field, the business knowledge base will match the problem. The business knowledge base (corresponding to the target knowledge base) has a huge amount of data. In order to improve efficiency, the business knowledge base classifies all business knowledge according to business. The classified service types include but are not limited to luggage, seat selection, meals, no-accompaniment and other additional services. The storage format of business knowledge in the knowledge base can be:

[0122]

[0123] When a user asks a business consultation question, business rules can be used to classify the input question first. For example, if the request entered by the user is related to luggage, similar questions will be found in the luggage classification questions, and the same is true for other classifications.

[0124] After clustering, the amount of data is still large, and the efficiency of directly using the model matching is still very slow. Therefore, the TFIDF algorithm can be used to perform a recall (similarity query) operation on the user input and the business questions in the specific classification, and the N questions with the highest similarity to the user questions (for example, ten questions) can be taken out, and then input into the Bert model to find the questions with the highest score similarity. If the similarity between the question and the question in the knowledge base is less than 50%, the user can be replied "The question cannot be found and can be transferred to manual processing". In other cases, the answer to the question is returned to the user.

[0125] For requests that cannot find similar problems, the operation module can collect them regularly and distribute them to professional operation personnel to perform business analysis operations and re-label the data to upgrade the model.

[0126] Step 7. User feedback: After the complete conversation between the system and the target object, the user can be invited to perform feedback operations and record the feedback questions. For the evaluation of user feedback, input it into the Bert model to perform user sentiment classification and analyze whether the user's evaluation is positive or negative. The operation module can regularly collect negative evaluations to improve system availability.

[0127] Step 8. Operation and maintenance: The operation module can regularly collect system logs, including but not limited to incorrectly classified input categories, negative user comments, and user inquiries that fail to find similar problems, and provide feedback to professional operation personnel. After analysis, operation personnel can annotate data for input requests that fail to be automatically classified; add business rules for negative user comments or add data annotations to input requests; and add questions and replies to the knowledge base for inquiries that fail to find similar problems.

[0128] The newly labeled data can be used to regularly fine-tune the Bert model to continuously optimize the model and upgrade the system to make the model better fit user needs.

[0129] Through this embodiment, users no longer need to fill out forms, but can complete inquiries and reservations through dialogues and natural language. The user input can be automatically classified based on the deep learning model trained and fine-tuned by BERT to automatically classify user needs, and inquire or book additional services or business consultation according to the classification results. Only a small number of unrecognizable user needs can be redirected to the manual customer service system for resolution, which can respond to user needs in real time and solve the problem of insufficient manpower. The additional service query system and the additional service reservation system can actively initiate dialogues to clarify the information required for reservation and query operations. For problems encountered during the reservation process, customers can get replies at any time by asking. After the reservation or query is completed, the user can give feedback on the service evaluation. The operation module will regularly collect feedback, input the deep learning model for sentiment classification, classify and collect the negative comments, and distribute them to the operation personnel. After the analysis and data marking by the operation personnel, the training of the Bert model can be implemented to promote system improvement and perfection.

[0130] Embodiment 3

[0131] Embodiment 3 of the present invention provides an optional service data processing device based on a deep learning model. Each implementation unit in the service data processing device corresponds to each implementation step in Embodiment 1.

[0132] Figure 4 is a schematic diagram of an optional service data processing device based on a deep learning model according to an embodiment of the present invention, such as Figure 4 As shown, the service data processing device based on the deep learning model includes: a first processing unit 41, a generating unit 42 and a second processing unit 43.

[0133] The first processing unit 41 is used to receive service request data input by the target object, identify the service type of the service request data, and obtain the target service type, wherein the service type of the service request data includes one of the following: a consulting class for consulting a service product, a query class for querying data of the service product, and a reservation class for reserving a service product;

[0134] The generating unit 42 is used to generate first reply data based on the target knowledge base and the service request data when the target service type is consulting, wherein the target knowledge base at least includes: M questions associated with the service product and the reply content of each question, and the first reply data is used to reply to the question carried in the service request data, and M is a positive integer:

[0135] The second processing unit 43 is used to input the service request data into the target deep learning model to obtain intention data when the target service type is a query type or a reservation type, and generate second reply data based on the intention data and the service request data, wherein the target deep learning model is used to identify the intention of the target object, and the intention of the target object includes one of the following: the need to perform a data query on a specified service product, and the need to reserve a specified service product, and the second reply data includes one of the following: a query result of a data query on a specified service product, and a reservation result of a reservation for a specified service product.

[0136] In the service data processing device based on the deep learning model provided in the third embodiment of the present invention, a first processing unit 41 can be used to receive the service request data input by the target object, and identify the service type of the service request data to obtain the target service type, wherein the service type of the service request data includes one of the following: a consultation class for consulting service products, a query class for querying data of service products, and a reservation class for reserving service products; a generation unit 42 is used to generate first reply data based on the target knowledge base and the service request data when the target service type is a consultation class, wherein the target knowledge base at least includes: M questions associated with the service product and the reply content of each question The first reply data is used to reply to the question carried in the service request data, and M is a positive integer: the second processing unit 43 is used to input the service request data into the target deep learning model to obtain the intention data when the target service type is the query type or the reservation type, and generate the second reply data based on the intention data and the service request data, wherein the target deep learning model is used to identify the intention of the target object, and the intention of the target object includes one of the following: the need to query the data of the specified service product, the need to reserve the specified service product, and the second reply data includes one of the following: the query result of the data query of the specified service product, the reservation result of the reservation of the specified service product. This solves the technical problem of poor processing effect of the request data of the additional service in the ticket booking process in the related technology.

[0137] In this embodiment, after receiving the service request data, the service request data of the additional service is automatically processed based on the service type and the deep learning model and corresponding reply data is generated, thereby avoiding the low processing efficiency of processing the service request data of the additional service by filling out forms or manual service in the related technology, thereby achieving the technical effect of improving the processing efficiency of the service request data of the additional service.

[0138] Optionally, in the service data processing device based on the deep learning model provided in Example 3 of the present invention, the generation unit includes: a first determination subunit, used to determine the similarity between the questions carried in the service request data and each question in the target knowledge base based on the word frequency-inverse document frequency algorithm, and obtain a similarity set; an extraction subunit, used to extract N target questions in the target knowledge base based on the similarity set, wherein the similarity between the target question and the question carried in the service request data is greater than the similarity between the questions other than the target question in the target knowledge base and the questions carried in the service request data, and N is a positive integer less than M; a generation subunit, used to generate first reply data based on the N target questions.

[0139] Optionally, in the service data processing device based on the deep learning model provided in Example 3 of the present invention, the generation subunit includes: a first processing module, used to input N target questions and service request data into the first deep learning model, output the score value of each target question, and determine the highest score value of the N target questions, wherein the first deep learning model is used to identify the score value of the similarity between the target question and the question carried in the service request data, and the score value of each target question is used to indicate the size of the similarity between the target question and the question carried in the service request data; a generation module, used to generate first reply data based on the reply content corresponding to the target question associated with the highest score value when the highest score value of the N target questions is greater than a preset threshold.

[0140] Optionally, in the service data processing device based on the deep learning model provided in Example 3 of the present invention, the first determination subunit includes: a second processing module, used to perform word segmentation processing on the questions carried in the service request data to obtain a word segmentation result; a third processing module, used to perform vectorization processing on the word segmentation result based on the word frequency-inverse document frequency algorithm to obtain a vectorized word segmentation result; a calculation module, used to calculate the cosine similarity between the vectorized word segmentation result and each question in the target knowledge base to obtain a similarity set.

[0141] Optionally, in the service data processing device based on the deep learning model provided in Example 3 of the present invention, the second processing unit includes: a second determination sub-unit, used to determine a preset slot based on intent data, wherein the preset slot includes: attributes of a field required to perform a target operation, and the target operation includes one of the following: data query for a specified service product, reservation for a specified service product; a filling sub-unit, used to fill the slot value of the preset slot based on the service request data to obtain a first slot filling result, wherein the slot value of the preset slot represents the attribute value of the attribute of the field required to perform the target operation; a first processing sub-unit, used to execute the target operation based on the first slot filling result, obtain an execution result, and generate second reply data based on the execution result, wherein the data format of the execution result is different from the data format of the second reply data.

[0142] Optionally, in the service data processing device based on the deep learning model provided in Example 3 of the present invention, the processing sub-unit includes: a detection module, used to detect whether there is a missing slot value in the preset slot to obtain a detection result; a first execution module, used to perform a target operation based on a first slot filling result when the detection result indicates that there is no missing slot value in the preset slot; a fourth processing module, used to obtain the missing slot value in the preset slot by conducting multiple rounds of dialogues with the target object when the detection result indicates that there is a missing slot value in the preset slot, obtain a target slot value, and fill the missing slot value in the preset slot based on the target slot value to obtain a second slot filling result; a second execution module, used to perform the target operation based on the second slot filling result.

[0143] Optionally, in the service data processing device based on the deep learning model provided in Example 3 of the present invention, the first processing unit includes: an identification subunit, used to identify the service type of the service request data based on preset rules to obtain an identification result; a second processing subunit, used to input the service request data into a second deep learning model and output a target service type when the identification result indicates that the service type identification of the service request data has failed, wherein the second deep learning model is used to identify the service type of the service request data; and a third determination subunit, used to determine the target service type based on the identification result when the identification result indicates that the service type identification of the service request data has been successful.

[0144] Optionally, in the service data processing device based on the deep learning model provided in Example 3 of the present invention, the service data processing device based on the deep learning model also includes: a third processing unit, which is used to obtain evaluation data of the target object after generating first reply data based on the target knowledge base, or after generating second reply data based on the intention of the target object, and determine the satisfaction of the target object based on the evaluation data.

[0145] The above-mentioned service data processing device based on deep learning model may also include a processor and a memory. The above-mentioned first processing unit 41, generation unit 42 and second processing unit 43, etc. are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize corresponding functions.

[0146] The above processor includes a kernel, and the kernel retrieves the corresponding program unit from the memory. One or more kernels can be set, and after receiving the service request data, the service request data of the additional service is automatically processed based on the service type and the deep learning model by adjusting the kernel parameters, and the corresponding reply data is generated, thereby avoiding the low processing efficiency of processing the service request data of the additional service by filling in a form or manual service in the related technology, thereby achieving the technical effect of improving the processing efficiency of the service request data of the additional service.

[0147] The above-mentioned memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one storage chip.

[0148] According to another aspect of an embodiment of the present invention, an electronic device is also provided, including: 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 service data processing methods based on a deep learning model by executing the executable instructions.

[0149] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is provided, which stores a computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the above-mentioned service data processing methods based on deep learning models.

[0150] Figure 5 is a schematic diagram of an electronic device according to an embodiment of the present invention, such as Figure 5 As shown, an embodiment of the present invention provides an electronic device 50, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, any one of the above-mentioned service data processing methods based on the deep learning model is implemented.

[0151] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0152] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0153] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0154] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0155] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0156] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0157] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A service data processing method based on deep learning model, It is characterized in that include: receiving service request data input by a target object, identifying a service type of the service request data, and obtaining a target service type, wherein the service type of the service request data includes one of the following: a consultation class for consulting a service product, a query class for querying data of the service product, and a reservation class for reserving the service product; In the case where the target service type is the consulting type, first reply data is generated based on the target knowledge base and the service request data, wherein the target knowledge base at least includes: M questions associated with the service product and the reply content of each question, and the first reply data is used to reply to the questions carried in the service request data, and M is a positive integer: When the target service type is the query class or the reservation class, the service request data is input into the target deep learning model to obtain intent data, and second reply data is generated based on the intent data and the service request data, wherein the target deep learning model is used to identify the intention of the target object, and the intention of the target object includes one of the following: the need to perform a data query on a specified service product, and the need to reserve the specified service product, and the second reply data includes one of the following: a query result of a data query on the specified service product, and a reservation result of a reservation for the specified service product.

2. The service data processing method according to claim 1, It is characterized in that Based on the target knowledge base and the service request data, generating first reply data includes: Based on a term frequency-inverse document frequency algorithm, determining the similarity between the question carried in the service request data and each question in the target knowledge base, and obtaining a similarity set; Based on the similarity set, extract N target questions in the target knowledge base, wherein the similarity between the target question and the question carried in the service request data is greater than the similarity between the questions other than the target question in the target knowledge base and the question carried in the service request data, and N is a positive integer less than M; Based on the N target questions, the first response data is generated.

3. The service data processing method according to claim 2, It is characterized in that Based on the N target questions, generating the first response data includes: Inputting N target questions and the service request data into a first deep learning model, outputting a score value for each target question, and determining the highest score value of the N target questions, wherein the first deep learning model is used to identify a score value of a similarity between the target question and the question carried in the service request data, and the score value of each target question is used to indicate the magnitude of the similarity between the target question and the question carried in the service request data; When the highest score value of the N target questions is greater than a preset threshold, the first reply data is generated based on the reply content corresponding to the target question associated with the highest score value.

4. The service data processing method according to claim 3, It is characterized in that Based on the term frequency-inverse document frequency algorithm, the similarity between the question carried in the service request data and each question in the target knowledge base is determined to obtain a similarity set, including: Performing word segmentation processing on the question carried in the service request data to obtain a word segmentation result; Based on the word frequency-inverse document frequency algorithm, the word segmentation result is vectorized to obtain a vectorized word segmentation result; The cosine similarity between the vectorized word segmentation result and each question in the target knowledge base is calculated to obtain a similarity set.

5. The service data processing method according to claim 1, It is characterized in that Generating second reply data based on the intention data and the service request data includes: Based on the intention data, a preset slot is determined, wherein the preset slot includes: an attribute of a field required to perform a target operation, and the target operation includes one of the following: performing a data query on the specified service product, and reserving the specified service product; Filling the slot value of the preset slot based on the service request data to obtain a first slot filling result, wherein the slot value of the preset slot represents an attribute value of an attribute of a field required to perform a target operation; Based on the first slot filling result, the target operation is executed to obtain an execution result, and the second reply data is generated based on the execution result, wherein the data format of the execution result is different from the data format of the second reply data.

6. The service data processing method according to claim 5, It is characterized in that Based on the first slot filling result, performing the target operation includes: Detecting whether there is a missing slot value in the preset slot, and obtaining a detection result; In a case where the detection result indicates that there is no missing slot value in the preset slot, based on the first slot filling result, performing the target operation; In the case where the detection result indicates that there is a missing slot value in the preset slot, a method of conducting multiple rounds of dialogues with the target object is adopted to obtain the missing slot value in the preset slot, obtain the target slot value, and fill the missing slot value in the preset slot based on the target slot value to obtain a second slot filling result; Based on the second slot filling result, perform the target operation.

7. The service data processing method according to claim 1, It is characterized in that Identifying the service type of the service request data and obtaining the target service type includes: Identify the service type of the service request data based on a preset rule to obtain an identification result; In a case where the recognition result indicates that the service type recognition of the service request data fails, the service request data is input into a second deep learning model, and the target service type is output, wherein the second deep learning model is used to recognize the service type of the service request data: In a case where the identification result indicates that the service type of the service request data is successfully identified, the target service type is determined based on the identification result.

8. The service data processing method according to claim 1, It is characterized in that After generating the first reply data based on the target knowledge base, or after generating the second reply data based on the intention of the target object, it also includes: acquiring the evaluation data of the target object, and determining the satisfaction of the target object based on the evaluation data.

9. A service data processing device based on a deep learning model, It is characterized in that include: A first processing unit is configured to receive service request data input by a target object, identify a service type of the service request data, and obtain a target service type, wherein the service type of the service request data includes one of the following: a consultation class for consulting a service product, a query class for querying data of the service product, and a reservation class for reserving the service product; A generating unit is used to generate first reply data based on a target knowledge base and the service request data when the target service type is the consulting type, wherein the target knowledge base at least includes: M questions associated with the service product and reply content of each question, and the first reply data is used to reply to the question carried in the service request data, and M is a positive integer: A second processing unit is used to input the service request data into a target deep learning model to obtain intent data when the target service type is the query class or the reservation class, and generate second reply data based on the intent data and the service request data, wherein the target deep learning model is used to identify the intention of the target object, and the intention of the target object includes one of the following: the need to perform a data query on a specified service product, and the need to reserve the specified service product, and the second reply data includes one of the following: a query result of a data query on the specified service product, and a reservation result of a reservation for the specified service product.

10. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores a computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the service data processing method based on the deep learning model as described in any one of claims 1 to 8.

11. An electronic device, It is characterized in that It includes one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the service data processing method based on the deep learning model as described in any one of claims 1 to 8.

Images