An intention recognition and broadcast method based on highway road conditions

Abstract

The application provides a kind of based on expressway road condition's intention recognition and broadcast method, including the following steps: collection expressway seat system incoming telephone, and classification is carried out according to consultation service;According to the form classification identification of key-intention, the intention classification model is generated;Collect the recording file of consultation road condition, and the consultation information of road condition information and different type structure is classified, and two kinds of data are marked as structure-type;Extract the key data in each road section monitoring reported event, then according to structure-type format is stored;According to the statement of incoming call user, the intention of user is judged, then the structure-type matched with the intention of incoming call user is selected from road condition information, then the matched information is converted into intelligent voice by using TTS technology, and broadcast to user.The application provides a kind of based on expressway road condition's intention recognition and broadcast method, can fully consider the consultation intention of incoming call user, let user obtain road condition information in first time, use more convenient.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of automobile driving assistance, and particularly relates to an intention recognition and broadcast method based on expressway road conditions. BACKGROUND

[0002] In the daily operation process of the expressway, expressway emergency information is usually only used for internal management of the expressway management department, and cannot provide services for the public traveling on the expressway in real time and conveniently. The public often travel first and then know, and cannot obtain the expressway road conditions in the first time to plan the trip. In order to improve the daily traffic rate and the public satisfaction rate of the expressway, the application combines NLP and TTS technologies to propose an intention recognition and broadcast method for expressway road conditions.

[0003] NLP (Neuro-Linguistic Programming) refers to neuro-linguistic programming (mind-body grammar programming), that is, the habits in our thinking and behavior, just like the programs in the computer, which can be changed by updating the software.

[0004] TTS is a kind of speech synthesis application, which converts the files stored in the computer, such as help files or web pages, into natural speech output. TTS can help people with visual impairment to read information on the computer, or simply to increase the readability of the text document. The current TTS application package includes voice-driven mail and sound-sensitive systems, and TTS is often used together with voice recognition programs. SUMMARY

[0005] Therefore, the application aims to provide an intention recognition and broadcast method based on expressway road conditions. The application aims to solve the problem that the public cannot obtain the expressway road conditions in the first time when driving and traveling, and the trip planning is inconvenient.

[0006] To achieve the above-mentioned purpose, the application provides an intention recognition and broadcast method based on expressway road conditions, which comprises the following steps:

[0007] S1. Collect the incoming calls of the expressway seat system in the past period of time, and divide the expressway business consultation into accident alarm, accident help, road condition consultation and ETC consultation according to the seat condition;

[0008] S2. Classify and identify the intention of the consultation business in the form of key-intention, generate an intention classification model based on the bert model, the key is the business type of consultation, and the intention is the intention recognition corresponding to the consultation business;

[0009] S3. Collecting the recorded files of the road condition inquiries in the past period of time in the expressway consultation telephone, classifying the inquiry information of the road condition in the recorded files into main line, service area, toll station and tunnel, dividing the inquiry information of different types of structures into congestion, accident and slow walking, and marking the two types of data as structure-type, the structure being the structure type identifier, and the type being the classification of the inquiry information;

[0010] S4. Extracting the key data in the events reported by each road section monitoring center, and then storing and recording the key data in the order of the occurrence of the events in the structure-type format;

[0011] S5. According to the statement of the incoming user, determining the intention of the incoming user, then selecting the structure-type matching the intention of the incoming user from the information reported by the expressway, and then converting the matched information into intelligent voice by using the TTS technology and broadcasting the intelligent voice to the user.

[0012] Further, the period of time in the step S1 and the step S3 is 6 months.

[0013] Further, in the step S3, the key data in the event includes the location of the occurrence of the event and the type of the event.

[0014] Further, in the step S5, the determination of the intention of the incoming user has the following two cases:

[0015] a. The intention of the user is within the range of the intention classification model in the step S2: determining the intention of the user by the intention classification model;

[0016] b. The intention of the user is outside the range of the intention classification model in the step S2: converting the statement of the user into a sentence vector, then performing dot multiplication on the sentence vector of the user and the sentence vector of the corpus calculated by the intention classification model to obtain the calculated intention;

[0017] If the calculated intention is similar to the content in the intention classification model and the similarity is greater than a certain threshold, the intention of the user can be determined by the intention classification model;

[0018] If the calculated intention is a new intention outside the range of the intention classification model, the intention of the user is directly output to the artificial service.

[0019] Further, the intention classification model in the step S2 is established by the Attention model, and the calculation steps of the Attention are as follows:

[0020] First step: calculating the weight by similarity between query and each key, the query representing the query of the incoming user, and the calculation expression of the weight being as follows:

[0021]

[0022] In the formula, Q represents query; K represents the i-th key; T represents the transpose of a matrix, which is an operation rule of the matrix; W represents the weight calculated in the first step; and V represents the input key value. i a a a All are learnable parameters.

[0023] The second step is to normalize the weight calculated in the first step by using a softmax function, and the calculation expression is as follows.

[0024]

[0025] The third step is to obtain attention by weighted sum of the weight and the corresponding key value value, the key = value, and the calculation expression of the attention is as follows.

[0026] Attention (Q, K, V) = ∑ i a i V i

[0027] In the formula, Q represents query; K represents key; V represents the input key value; and a represents the weight calculated in the second step. i i

[0028] Further, the incoming call user intention discrimination also adopts the Attention model, and the judgment process is as follows: the key value of the user voice is recognized, and then calculation and discrimination are performed through the Attention model.

[0029] The present application has the following advantages:

[0030] 1. The present application provides an intention recognition and broadcast method based on highway road conditions, combines NLP and TTS technologies, establishes a prefabricated intention classification model and structure-type data structure, fully considers the incoming call user consultation information range, and makes an early prediction, so that feedback can be made at the first time after receiving the consultation; in the input process of the statement of the incoming call user and the output process of the intelligent voice, key structure-type data are extracted for matching, and the information fed back to the user is faster and more accurate.

[0031] ​​​​​2.The application provides a highway traffic condition-based intention recognition and broadcast method, which uses an Attention model to pay attention to and fully learn and absorb collected information, and the Attention model is more accurate in output and less likely to lose text information compared with a traditional Seq2Seq model.

[0032] Other advantages, objects, and features of the application will be set forth in part in the following specification taken in conjunction with the accompanying drawings, and in part will become apparent to those skilled in the art from a consideration of the following specification and drawings. The objects and other advantages of the application will be realized and attained by means of the instrumentalities and combinations pointed out in the following specification. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 A flowchart of the highway traffic condition-based intention recognition and broadcast method of the application is shown in the figure.

[0034] Figure 2 The figure is a flowchart of the attention function calculation of the application. DETAILED DESCRIPTION

[0035] In order to make the technical solutions, advantages and purposes of the application clearer, the technical solutions of the embodiments of the application will be described clearly and completely below in conjunction with the drawings of the embodiments of the application. Obviously, the described embodiments are part of the embodiments of the application, rather than all the embodiments. Based on the described embodiments of the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the present application.

[0036] The application provides a highway traffic condition-based intention recognition and broadcast method, which comprises the following steps:

[0037] S1.Collect the incoming calls of the highway call center system in the past 6 months, and combine the actual situation of the existing call center to divide the highway service consultation into accident alarm, accident help, traffic consultation and ETC consultation in detail.

[0038] S2.Classify and identify the intention of each consultation service in the form of key-intention to generate an intention classification model based on the bert model, wherein key represents the type of consultation service, and intention represents the intention recognition corresponding to the consultation service.

[0039] BERT is a language representation model, BERT stands for Bidirectional Encoder Representations from Transformers. BERT aims to pretrain deep bidirectional representations through jointly conditioning on both left and right context in all layers. As a result, a single additional output layer is needed to adapt pre-trained BERT representations for a wide range of tasks, such as question answering and language inference tasks, creating state-of-the-art models without the need for substantial model-specific modifications.

[0040] S3. Collect the audio files of the road condition inquiries in the daily consultation calls of the expressway in the past 6 months, classify the inquiry road condition information in the audio files into main line, service area, toll station and tunnel, classify the inquiry information of different types of structures into congestion, accident and slow walking, and mark the two types of data as structure-type, wherein structure represents the structure type identifier, and type represents the information classification of the inquiry.

[0041] S4. Extract the key data in the event information reported by the section monitoring center, the key information includes the location of the event and the type of the event, and then store and record the events in the order of occurrence according to the structure-type format;

[0042] S5. According to the statement of the incoming call user, the intention of the incoming call user is judged, then the structure-type of the statement entering the flow control module is extracted, then the structure-type matching the intention of the incoming call user is selected from the information reported by the expressway, and then the matching information is converted into intelligent voice by using TTS technology and broadcast to the user.

[0043] As a preferred embodiment of the present embodiment, in step S5, the judgment of the intention of the incoming call user has the following two cases:

[0044] a. When the intention of the incoming call user is within the range of the intention classification model in step S2: the intention of the user is judged by the intention classification model;

[0045] b. When the intention of the incoming call user is outside the range of the intention classification model in step S2: the sentence vector of the user is converted by using the bert model, then the dot product of the sentence vector of the user and the sentence vector of the corpus calculated by the intention classification model is calculated, and the most likely intention after calculation is obtained;

[0046] If the calculated intention is similar to the content in the intention classification model, and the similarity is greater than a certain threshold, the intention of the user can be judged by the intention classification model;

[0047] If the calculated intention is a new intention, which is out of the range of the intention classification model, the user intention is directly output to the manual service.

[0048] As a preferred embodiment of the present embodiment, the intention classification model in step S2 is established by an Attention model. The model focuses on important information and learns and absorbs it fully through the attention mechanism. The calculation steps of Attention are as follows:

[0049] First step: similarity calculation between query and each key to obtain weight. Query represents the query of the incoming user, and the functions for calculating similarity include dot product, splicing, and perception machine. The calculation expression of weight is as follows:

[0050]

[0051] In the formula, Q represents query; K represents the i-th key; T represents the transpose of the matrix, which is a matrix operation rule; W, v, and U are all learnable parameters. i a a a

[0052] Second step: using the softmax function to normalize the weight calculated in the first step. The calculation expression is as follows:

[0053]

[0054] Third step: weighted summation of weight and corresponding key value value to obtain attention. Key = value. The calculation expression of attention is as follows:

[0055] Attention(Q, K, V) = å i i i

[0056] In the formula, Q represents query; K represents key; V represents the input key value; and a represents the weight calculated in the second step. i i

[0057] As a preferred embodiment of the present embodiment, the identification of the intention of the incoming user also adopts the Attention model. The judgment process is as follows: the key value of the real-time incoming user voice is recognized, then the Attention model is calculated, the closest intention classification is matched, if it is a classification that has been pre-set in the intention classification model, the subsequent process is executed; if it is a new statement vector, the intention is directly output to the manual service.

[0058] ​​​​​​​​Finally, it is to be explained that the above embodiments are only used to illustrate the technical solutions of the present application but not to limit the present application. Although the present application is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or equivalently replaced without departing from the purpose and scope of the technical solutions, and all should be covered in the protection scope of the present application.

Claims

1. A highway traffic condition-based intention recognition and broadcast method, comprising the following steps: S1. Collecting incoming calls of the highway call center system in the past period of time, and dividing the highway service consultation into accident alarm, accident help, traffic condition consultation and ETC consultation according to the call center situation; S2. Classifying the intention of the consultation service in the form of key-intention, generating an intention classification model based on the bert model, wherein the key is the type of consultation service, and the intention is the intention recognition corresponding to the consultation service; The intention classification model in step S2 is established by an Attention model, and the calculation steps of the Attention are as follows: First step: similarity calculation is performed on the query and each key to obtain a weight, wherein the query represents the query of the incoming user, and the calculation expression of the weight is as follows: where Q represents query; K i represents the i-th key; T indicates the transpose of a matrix, which is an operation rule of a matrix; W a , v a , U a are all learnable parameters; Second step: using a softmax function to normalize the weight calculated in the first step, and the calculation expression is as follows; Third step: weight and corresponding key value value weighted sum of attention, the key = value, the attention calculation expression as follows: where Q represents query; K represents key; V i is the input key value; a i represents the weight calculated in the second step; S3. Collecting the audio files of the road condition consultation in the past period of time, classifying the road condition information in the audio files into main line, service area, toll station and tunnel, classifying the consultation information of different types of structures into congestion, accident and slow walking, and marking the two types of data as structure-type, wherein the structure is a structure type identifier, and the type is an information classification of the consultation; S4. Extracting the key data in the events reported by the road section monitoring center, and then storing and recording the events in the order of occurrence according to the structure-type format; S5. According to the statement of the incoming user, the intention of the incoming user is judged, and then the structure-type matching the intention of the incoming user is selected from the information reported by the highway, and then the matching information is converted into intelligent voice by using TTS technology and broadcast to the user.

2. The method of claim 1, wherein the method is based on highway traffic. The period of time in steps S1 and S3 is 6 months. 3.The highway traffic-based intention recognition and broadcasting method of claim 1, wherein: In step S3, the key data in the event includes the location of the event and the type of the event. 4.The highway traffic-based intention recognition and broadcasting method of claim 1, wherein, In step S5, the judgment of the intention of the incoming user has the following two cases: a. The user's intention is within the range of the intention classification model in step S2: the user's intention is judged by the intention classification model; b. The user's intention is outside the range of the intention classification model in step S2: the user's sentence is converted into a sentence vector, and then the dot product of the user's sentence vector and the sentence vector of the corpus calculated by the intention classification model is calculated to obtain the calculated intention; If the calculated intention is similar to the content in the intention classification model and the similarity is greater than a certain threshold, the user's intention can be judged by the intention classification model; If the calculated intention is a new intention outside the range of the intention classification model, the user's intention is directly output to manual service. 5.The highway traffic-based intention recognition and announcement method of claim 4, wherein, The judgment of the intention of the incoming user also uses an Attention model, and the judgment process is as follows: The key value of the user's voice is recognized, and then calculated and judged by the Attention model.

Images