Information identification method, device, computer equipment and storage medium

By integrating the identification model of CNN and DBM models, combining the local and global characteristics of user text information, combined with the number of underwriting years and predicted probability, the accurate identification of user needs is achieved, and the problems of inefficiency and high labor costs in the existing technology are solved.

CN111414451BActive Publication Date: 2024-11-08CHINA PING AN PROPERTY INSURANCE CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202010124754.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-02-27
Publication Date
2024-11-08
Estimated Expiration
2040-02-27

AI Technical Summary

Technical Problem

The prior art cannot accurately identify user needs, resulting in inefficiency and high labor costs, and the subjectivity of different seats leads to inconsistent identification results.

Method used

The recognition model that combines the CNN model and the DBM model is adopted to obtain user text information through speech conversion, extract local and global feature vectors for fusion, perform classification calculations, and weighted calculations based on the number of underwriting years and predicted probability to achieve accurate identification of user needs.

Benefits of technology

It improves the accuracy of user demand identification, reduces labor costs, avoids the inaccuracy of subjective identification, and achieves more efficient information interaction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN111414451B_ABST
    Figure CN111414451B_ABST
Patent Text Reader

Abstract

The present application relates to the technical field of classification models, and provides an information recognition method, device, computer equipment and storage medium, which receives voice information returned by a target user, performs voice conversion on the voice information, and obtains corresponding user text information; the user text information is respectively input into a preset recognition model; the recognition model is obtained by training a fusion CNN model and a DBM model; the local feature vector of the user text information is extracted by the fully connected layer of the CNN model, and the global feature vector of the user text information is extracted by the DBM model; the local feature vector is then fused with the global feature vector to obtain a fused feature vector and input it into the classification layer; finally, the fused feature vector is classified and calculated by the classification layer to obtain the classification result corresponding to the user text information. The present application not only takes into account the local features in the user text information, but also combines the global features, so that the final classification result is more accurate.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of classification models, and in particular to an information identification method, apparatus, computer device and storage medium. Background Art

[0002] At present, due to the continuous increase in seat labor costs, various companies are in urgent need of reducing labor costs and will use intelligent robots to conduct information interaction for various products. In the process of information interaction with users, it is necessary to identify whether the user has corresponding needs. However, currently, user behavior can only be identified manually, and intelligent robots cannot be used to identify whether the user has corresponding needs from the user's answer information, resulting in low efficiency and high labor costs. At the same time, during manual identification, different seats have different subjectivities, which will make the identification results different, so it is impossible to accurately identify the user's true needs. Summary of the invention

[0003] The main purpose of this application is to provide an information identification method, apparatus, computer equipment and storage medium, aiming to solve the current defect that user needs cannot be accurately identified from the user's answer information.

[0004] To achieve the above purpose, the present application provides an information identification method, comprising the following steps:

[0005] Read the continuous insurance years of each user stored in the preset database, and select the target user whose continuous insurance years are greater than the preset value;

[0006] Outputting preset sales text information to the user terminal where the target user is located;

[0007] Receiving the voice information returned by the target user through the user terminal, and performing voice conversion on the voice information to obtain corresponding user text information;

[0008] The user text information is respectively input into a preset recognition model; the recognition model is obtained by training a fusion CNN model and a DBM model, wherein a classification layer connected to the fully connected layer of the CNN model and the feature output layer of the DBM model is used as the final output layer of the recognition model;

[0009] Extracting a local feature vector of the user text information by the fully connected layer of the CNN model, and extracting a global feature vector of the user text information by the DBM model;

[0010] Fusing the local feature vector with the global feature vector to obtain a fused feature vector and inputting the fused feature vector into the classification layer;

[0011] The fused feature vector is classified and calculated by the classification layer to obtain a classification result corresponding to the user text information as a recognition result of the voice information returned by the target user.

[0012] Furthermore, after the step of performing classification calculation on the fused feature vector through the classification layer to obtain the classification result corresponding to the user text information as the recognition result of the voice information returned by the target user, the method further comprises:

[0013] Create a blank layer;

[0014] Obtaining the number of characters of the sales text information and the user text information;

[0015] Calculate the first area size and the second area size occupied by the sales text information and the user text information when added to the blank layer according to the number of characters of the sales text information and the user text information and the preset font format;

[0016] Dividing the blank layer into a first area corresponding to the size of the first area and a second area corresponding to the size of the second area; and dividing a third area of ​​a preset size between the first area and the second area in the blank layer;

[0017] The sales text information is added into the first area, the user text information is added into the second area, the classification result is added into the third area, and they are combined into a picture.

[0018] Furthermore, the step of adding the sales text information into the first area, adding the user text information into the second area, and adding the classification result into the third area includes:

[0019] Performing hash calculations on the sales text information, the user text information, and the classification result to obtain corresponding sales hash values, user hash values, and classification hash values;

[0020] Encrypting the sales text information by using the classification hash value, and adding the encrypted sales text information into the first area;

[0021] Encrypting the user text information by using the sales hash value, and adding the encrypted user text information into the second area;

[0022] The classification result is encrypted by the user hash value, and the encrypted classification result is added to the third area.

[0023] Furthermore, the classification result includes an underwriting label and a corresponding prediction probability; after the step of performing classification calculation on the fused feature vector through the classification layer to obtain the classification result corresponding to the user text information as the recognition result of the voice information returned by the target user, the method further includes:

[0024] Saving the sales text information and user text information in the preset database;

[0025] Combining the sales text information with the underwriting label to obtain a first combination;

[0026] Combining the user text information with the predicted probability corresponding to the underwriting label to obtain a second combination;

[0027] Calculate the first combination and the second combination respectively by using a hash algorithm to obtain a corresponding first hash value and a second hash value;

[0028] The first hash value and the second hash value are added to the same array to generate a first collection; and the first collection is uploaded to the enterprise underwriting management server for storage.

[0029] Furthermore, before the step of reading the continuous insurance years of each user stored in the preset database and selecting the target user whose continuous insurance years are greater than the preset value, the step includes:

[0030] Based on the spark data computing engine, obtain the user's historical insurance policy underwriting data;

[0031] Calculate the user's historical insurance policy underwriting data to obtain the user's underwriting year;

[0032] Based on the array sort algorithm, the insurance years are sorted to obtain an insurance year sequence arranged from largest to smallest;

[0033] Calculating the differences between adjacent underwriting years in the underwriting year sequence in sequence, and sequentially composing the differences into a difference sequence;

[0034] Inputting the corresponding values ​​in the difference sequence into a preset formula to calculate the number of consecutive years of insurance coverage for the user;

[0035] The user's continuous insurance years are associated with the user's user information and stored in a preset database.

[0036] Furthermore, after the step of inputting the corresponding values ​​in the difference sequence into a preset formula to calculate the number of consecutive years of insurance coverage for the user, the method further comprises:

[0037] According to the insurance year of the user, obtain the number of years the user is insured;

[0038] According to the maximum year and the minimum year among the insured years of the user, obtain the year span of the user's insurance;

[0039] Calculate the user's continuous insurance probability according to the number of years the user has been insured, the number of consecutive years of insurance, and the year span;

[0040] The classification result includes an underwriting label and a corresponding prediction probability. After the step of performing classification calculation on the fused feature vector through the classification layer to obtain the classification result corresponding to the user text information as the recognition result of the voice information returned by the target user, the method includes:

[0041] The underwriting probability of the user is obtained by performing a weighted calculation based on the continuous underwriting probability and the predicted probability corresponding to the underwriting label.

[0042] The present application also provides an information identification device, comprising:

[0043] A selection unit is used to read the continuous insurance years of each user stored in the preset database, and select a target user whose continuous insurance years are greater than a preset value;

[0044] An output unit, used for outputting preset sales text information to a user terminal where the target user is located;

[0045] A conversion unit, configured to receive the voice information returned by the target user through the user terminal, and perform voice conversion on the voice information to obtain corresponding user text information;

[0046] An input unit, used to input the user text information into a preset recognition model respectively; the recognition model is obtained by training a fusion CNN model and a DBM model, wherein a classification layer connected to the fully connected layer of the CNN model and the feature output layer of the DBM model is used as the final output layer of the recognition model;

[0047] An extraction unit, configured to extract a local feature vector of the user text information through the fully connected layer of the CNN model, and extract a global feature vector of the user text information through the DBM model;

[0048] A fusion unit, used for fusing the local feature vector with the global feature vector to obtain a fused feature vector and input it into the classification layer;

[0049] The classification unit is used to perform classification calculation on the fused feature vector through the classification layer to obtain a classification result corresponding to the user text information as a recognition result of the voice information returned by the target user.

[0050] Furthermore, it also includes:

[0051] Create a unit to create a blank layer;

[0052] A first acquisition unit, used to acquire the number of characters of the sales text information and the user text information;

[0053] A first calculation unit is used to calculate the first area size and the second area size occupied by the sales text information and the user text information when added to the blank layer according to the number of characters of the sales text information and the user text information and a preset font format;

[0054] A division unit, configured to divide the blank layer into a first area corresponding to the size of the first area and a second area corresponding to the size of the second area; and to divide a third area of ​​a preset size between the first area and the second area in the blank layer;

[0055] The adding unit is used to add the sales text information into the first area, add the user text information into the second area, add the classification result into the third area, and synthesize them into a picture.

[0056] Furthermore, the adding unit comprises:

[0057] A calculation subunit, used to perform hash calculations on the sales text information, the user text information and the classification result, respectively, to obtain corresponding sales hash values, user hash values ​​and classification hash values;

[0058] A first adding subunit, configured to encrypt the sales text information by using the classification hash value, and add the encrypted sales text information into the first area;

[0059] a second adding subunit, configured to encrypt the user text information by using the sales hash value, and add the encrypted user text information into the second area;

[0060] The third adding subunit is used to encrypt the classification result by using the user hash value, and add the encrypted classification result to the third area.

[0061] Furthermore, the classification result includes an underwriting label and a corresponding predicted probability; the information identification device further includes:

[0062] A storage unit, used for storing the sales text information and the user text information in the preset database;

[0063] A first combining unit, used to combine the sales text information with the underwriting label to obtain a first combination;

[0064] A second combining unit, configured to combine the user text information with the predicted probability corresponding to the underwriting label to obtain a second combination;

[0065] A second calculation unit, configured to calculate the first combination and the second combination respectively by using a hash algorithm to obtain a corresponding first hash value and a second hash value;

[0066] An uploading unit is used to add the first hash value and the second hash value into the same array to generate a first collection; and upload the first collection to the enterprise underwriting management server for storage.

[0067] Furthermore, the information identification device further includes:

[0068] The second acquisition unit is used to acquire the user's historical insurance policy underwriting data based on the spark data computing engine;

[0069] A third calculation unit is used to calculate the user's historical insurance policy underwriting data to obtain the user's underwriting year;

[0070] A sorting unit, used for sorting the insurance years based on an array sort algorithm to obtain an insurance year sequence arranged from largest to smallest;

[0071] A fourth calculation unit is used to sequentially calculate the differences between adjacent underwriting years in the underwriting year sequence, and sequentially form the differences into a difference sequence;

[0072] A fifth calculation unit, used for inputting the corresponding values ​​in the difference sequence into a preset formula to calculate the continuous insurance years of the user;

[0073] The storage unit is used to associate the user's continuous insurance years with the user's user information and store them in a preset database.

[0074] Furthermore, the information identification device further includes:

[0075] A third acquisition unit, configured to acquire the number of years of insurance coverage for the user according to the insurance year of the user;

[0076] A fourth acquisition unit, configured to acquire the year span of the user's insurance according to the maximum year and the minimum year of the user's insurance years;

[0077] a sixth calculation unit, configured to calculate the continuous insurance probability of the user according to the number of years of insurance of the user, the number of consecutive insurance years and the year span;

[0078] The classification result includes an underwriting label and a corresponding predicted probability, and the information identification device further includes:

[0079] The seventh calculation unit is used to perform weighted calculation according to the continuous underwriting probability and the predicted probability corresponding to the underwriting label to obtain the underwriting probability of the user.

[0080] The present application also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0081] The present application also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0082] The information recognition method, device, computer equipment and storage medium provided in the present application receive the voice information returned by the target user, and perform voice conversion on the voice information to obtain the corresponding user text information; the user text information is respectively input into the preset recognition model; the recognition model is obtained by training the fusion CNN model and the DBM model; the local feature vector of the user text information is extracted by the fully connected layer of the CNN model, and the global feature vector of the user text information is extracted by the DBM model; the local feature vector is then fused with the global feature vector to obtain a fused feature vector and input it into the classification layer; finally, the fused feature vector is classified and calculated by the classification layer to obtain the classification result corresponding to the user text information as the recognition result of the voice information returned by the target user. The present application not only takes into account the local features in the user text information, but also combines the global features, so that the final classification result is more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 This is a schematic diagram of the steps of the information identification method in one embodiment of the present application;

[0084] Figure 2 This is a structural block diagram of an information identification device in an embodiment of the present application;

[0085] Figure 3 A schematic block diagram of the structure of a computer device according to an embodiment of the present application.

[0086] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0087] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0088] Reference Figure 1 In one embodiment of the present application, an information identification method is provided, comprising the following steps:

[0089] Step S1, reading the continuous insurance years of each user stored in a preset database, and selecting a target user whose continuous insurance years are greater than a preset value;

[0090] Step S2, outputting preset sales text information to the user terminal where the target user is located;

[0091] Step S3, receiving the voice information returned by the target user through the user terminal, and performing voice conversion on the voice information to obtain corresponding user text information;

[0092] Step S4, inputting the user text information into a preset recognition model respectively; the recognition model is obtained by training a fusion CNN model and a DBM model, wherein a classification layer connected to the fully connected layer of the CNN model and the feature output layer of the DBM model is used as the final output layer of the recognition model;

[0093] Step S5, extracting a local feature vector of the user text information by the fully connected layer of the CNN model, and extracting a global feature vector of the user text information by the DBM model;

[0094] Step S6, fusing the local feature vector with the global feature vector to obtain a fused feature vector and inputting it into the classification layer;

[0095] Step S7, performing classification calculation on the fused feature vector through the classification layer to obtain a classification result corresponding to the user text information as a recognition result of the voice information returned by the target user.

[0096] In this embodiment, the above-mentioned information identification method is applied to the scenario of obtaining the customer's underwriting needs. It is implemented based on an intelligent robot. The intelligent robot replaces manual labor to identify the customer's underwriting needs, which not only reduces labor costs, but also avoids the inaccurate identification caused by manual subjective consciousness.

[0097] In this embodiment, the above method is aimed at high-quality customers. As described in step S1 above, the high-quality customers in this embodiment refer to target users whose continuous insurance years are greater than a preset value; when the target user has a large number of continuous insurance years, it indicates that the probability of the user being insured is also correspondingly high, and therefore is defined as a high-quality customer.

[0098] As described in step S2 above, for the target user, the preset sales text information is output to the user terminal where the target user is located, so that the target user can view the corresponding sales text information, which includes the insurance content sold, discount information and other related information.

[0099] As described in the above step S3, the target user can make a corresponding reply to the above sales text information. For the convenience of the user, he / she can use voice to answer. After the user terminal where the target user is located collects the target user's voice information, it sends it to the above intelligent robot; the above intelligent robot can then convert the above voice information into user text information.

[0100] As described in step S4 above, a recognition model (CNN-DBM model) is pre-trained, which is a new model obtained by training based on the CNN (convolutional neural network) model and the DBM (data-based mechanistic) model. The recognition model integrates the CNN model and the DBM model, so that it has a better effect in classifying user text information and has a better ability to recognize professional terms and texts. Specifically, the recognition model integrates the CNN model and the DBM model, and the fully connected layer of the CNN model and the feature output layer of the DBM model are connected to a classification layer as the final output layer of the recognition model; that is, the recognition model in this embodiment not only includes the above-mentioned CNN model and the DBM model, but also no longer outputs the classification results separately by the above-mentioned CNN model and the DBM model, but the CNN model and the DBM model perform feature extraction respectively, and the extracted features are fused and then the classification results are output by the classification layer of the above-mentioned recognition model. In this embodiment, when training to obtain the above-mentioned recognition model, it is necessary to train the above-mentioned CNN model and DBM model separately, and then extract features from the CNN model and the DBM model respectively and input them into the classification layer for fusion training to obtain the above-mentioned recognition model.

[0101] As described in step S5 above, the feature vector of the user text information is extracted by the CNN model and the DBM model in the recognition model, respectively, wherein the CNN model extracts the corresponding feature vector through the last layer of its fully connected layer (i.e., the previous layer of the output layer). Since the CNN model extracts features by sliding the convolution kernel when extracting feature vectors, the feature vector extracted by it is a local feature vector. The DBM model is an undirected graph-connected neural network, which samples and calculates the nodes of the two connected layers together to make the text representation more accurate. Therefore, the DBM model extracts the global feature vector, and the DBM model extracts the global feature vector through its output layer. In the CNN model, the last layer of the fully connected layer outputs the local feature vector corresponding to 2 output categories (the classification result has two labels, the underwriting label and the non-underwriting label). The DBM model also corresponds to 2 output categories; the CNN model and the DBM model both extract the corresponding feature vector in each output category, and then fuse the local feature vector and the global feature vector of each output category.

[0102] As described in step S6 above, the local feature vector and the global feature vector are fused to output the corresponding fused feature vector. The feature vectors of the user text information extracted by the CNN model and the DBM model are different, and the target feature vector obtained after fusion has different characteristics of the feature vectors.

[0103] As described in step S7 above, in this embodiment, the fused feature vector is input into the classification layer for calculation, which is consistent with the calculation principle when a single feature vector is input, except that the input vector (here, the fused feature vector obtained by fusion) is different. The fused feature vector after the fusion of the above two feature vectors is a new vector, which is input into the classification layer for calculation and finally classified. The classification result incorporates the characteristics of different feature vectors, and the classification result obtained is more accurate than the recognition and classification when the above feature vectors are input separately. Therefore, the effect of identifying and classifying the insurance needs of user text information is better, and the user's real insurance needs can be identified more accurately.

[0104] In one embodiment, after the step S7 of performing classification calculation on the fused feature vector by the classification layer to obtain the classification result corresponding to the user text information as the recognition result of the voice information returned by the target user, the method includes:

[0105] Step S8, create a blank layer; the blank layer is used to add the above-mentioned sales text information, user text information and classification results, so as to solidify the information generated in the process and facilitate subsequent investigation.

[0106] Step S9, obtaining the number of characters in the sales text information and the user text information; the above number of characters is the number of characters in the above text information.

[0107] Step S10, according to the number of characters in the sales text information and the user text information, and the preset font format, calculate the first area size and the second area size occupied by the sales text information and the user text information added to the blank layer. The more characters there are and the larger the font is, the larger the area size will be. In this embodiment, the same preset font format is used when the sales text information and the user text information are added to the blank layer.

[0108] Step S11, dividing the blank layer into a first area corresponding to the size of the first area and a second area corresponding to the size of the second area; and dividing a third area of ​​a preset size between the first area and the second area in the blank layer.

[0109] Step S12, adding the sales text information into the first area, adding the user text information into the second area, adding the classification result into the third area, and synthesizing them into a picture.

[0110] In this embodiment, the sales text information and user text information are added to different areas of the blank layer, which is convenient for distinguishing the above information; at the same time, the above classification results are also added to the above blank layer, which is convenient for directly obtaining the above classification results when viewing later. In this embodiment, the above text information is solidified in the same layer to form a picture, which not only facilitates the recording of information and avoids data clutter, but also avoids data loss and tampering.

[0111] In one embodiment, the step S12 of adding the sales text information into the first area, adding the user text information into the second area, and adding the classification result into the third area includes:

[0112] Step S121, performing hash calculations on the sales text information, the user text information and the classification result respectively to obtain corresponding sales hash values, user hash values ​​and classification hash values;

[0113] Step S122, encrypting the sales text information by using the classification hash value, and adding the encrypted sales text information to the first area;

[0114] Step S123, encrypting the user text information by using the sales hash value, and adding the encrypted user text information to the second area;

[0115] Step S124: encrypt the classification result by using the user hash value, and add the encrypted classification result to the third area.

[0116] In this embodiment, the above-mentioned sales text information, user text information, and classification results usually contain sensitive information of the user. In order to prevent the above-mentioned sensitive information from being leaked, the above-mentioned sales text information, user text information, and classification results need to be encrypted.

[0117] In this embodiment, the encryption method adopted is to perform hash calculations on the sales text information, the user text information and the classification result respectively to obtain corresponding sales hash values, user hash values ​​and classification hash values; encrypt the sales text information by the classification hash value, encrypt the user text information by the sales hash value, and encrypt the classification result by the user hash value. The sales text information, user text information and classification result are encrypted by this interlocking encryption method, which has the following advantages: first, the encryption method is novel, and the encryption of the next information needs to be combined with the hash value characteristics of the previous information; second, when saving the password, only any one of the encrypted passwords needs to be saved, which reduces the difficulty of remembering the password; after decrypting one of the information by the encrypted password, all the information can be obtained in sequence. When other people do not know the arrangement order and decryption rules of the above information, even if they know one of the encrypted passwords, they cannot continue to decrypt the password of the next information.

[0118] In one embodiment, the classification result includes an underwriting label and a corresponding prediction probability; after the classification calculation of the fused feature vector by the classification layer to obtain the classification result corresponding to the user text information as the recognition result of the voice information returned by the target user, the following steps are included:

[0119] Step S71, saving the sales text information and user text information in the preset database;

[0120] Step S72, combining the sales text information with the insurance label to obtain a first combination;

[0121] Step S73, combining the user text information with the predicted probability corresponding to the underwriting label to obtain a second combination;

[0122] Step S74, respectively calculating the first combination and the second combination by a hash algorithm to obtain a corresponding first hash value and a second hash value;

[0123] Step S75, adding the first hash value and the second hash value to the same array to generate a first collection; uploading the first collection to the enterprise underwriting management server for storage.

[0124] In this embodiment, in order to facilitate the subsequent review and investigation of the above-mentioned sales text information and user text information, and to prevent users from tampering with the above-mentioned information, it is necessary to save the above-mentioned sales text information and user text information on the enterprise underwriting management server; however, if the above-mentioned sales text information and user text information are all saved on the above-mentioned enterprise underwriting management server, it will undoubtedly cause the storage cost of the enterprise underwriting management server to increase, thereby increasing the pressure on the server.

[0125] Therefore, in this embodiment, the above-mentioned sales text information and user text information are only stored in a local preset database, which can reduce the storage pressure of the above-mentioned enterprise underwriting management server; and in order to prevent the above-mentioned sales text information and user text information from being tampered with, in this embodiment, the sales text information is combined with the underwriting label to obtain a first combination, and the user text information is combined with the predicted probability corresponding to the underwriting label to obtain a second combination; then, the first combination and the second combination are calculated respectively by a hash algorithm to obtain the corresponding first hash value and second hash value; finally, the first hash value The second hash value is packaged into a first collection; the first collection is uploaded to the enterprise underwriting management server for storage; only a collection is stored on the enterprise underwriting management server, which does not increase too much storage pressure; and the advantage of the above collection is that when it is necessary to check the above locally stored sales text information and user text information and verify whether the above information is true, it is only necessary to regenerate a second collection according to the above method; and compare the second collection with the first collection stored on the enterprise underwriting management server. If they are consistent, it indicates that the information has not been tampered with and is true information; if they are inconsistent, it indicates that the information has been tampered with.

[0126] In one embodiment, before the step S1 of reading the continuous insurance years of each user stored in the preset database and selecting the target user whose continuous insurance years are greater than the preset value, the following steps are included:

[0127] Step S101, based on the spark data computing engine, obtain the user's historical insurance policy underwriting data;

[0128] Step S102, calculating the user's historical insurance policy underwriting data to obtain the user's underwriting year;

[0129] Step S103, sorting the insurance years based on an array sort algorithm to obtain an insurance year sequence arranged from largest to smallest;

[0130] Step S104, sequentially calculating the differences between adjacent underwriting years in the underwriting year sequence, and sequentially composing the differences into a difference sequence;

[0131] Step S105, inputting the corresponding values ​​in the difference sequence into a preset formula to calculate the number of consecutive years of insurance coverage for the user;

[0132] Step S106: associate the user's continuous insurance years with the user's user information and store them in a preset database.

[0133] Among them, in the above step S101, the user's historical insurance policy underwriting data refers to the insurance policy data related to the user's underwriting during the insurance period, such as historical underwriting years or specific underwriting time data. Specifically, the user's historical insurance policy underwriting data can be obtained from the background database of the insurance system, or from the insurance web page through crawler technology, or through the data collection interface to collect the user's historical insurance policy underwriting data or through the big data platform. Preferably, in this embodiment, the Spark computing engine of the big data platform is used to obtain the user's historical insurance policy underwriting data. Since the user's historical insurance policy underwriting data contains massive data, it is more convenient and efficient to use the big data platform to obtain data, so as to facilitate the subsequent further processing of the user's historical insurance policy underwriting data. Among them, Spark is an open source cluster computing environment. Spark enables memory distributed data sets. In addition to providing interactive queries, it can also optimize iterative workloads.

[0134] Specifically, all insurance policies corresponding to the vehicle ID are queried from the user's historical insurance policy claim data. It is understandable that due to the massive amount of user historical insurance policy claim data, the calculation amount of consecutive insurance years is also large and they are all repeated calculations. Therefore, through the iterative optimization calculation of the big data computing engine spark, the consecutive insurance years can be calculated quickly and accurately.

[0135] In the above step S102, the underwriting year refers to the year corresponding to the insurance policy transaction time. For example, if user A is insured in 2014, 2016, 2017, 2018, and 2019, then the underwriting years are 2014, 2016, 2017, 2018, and 2019.

[0136] In the above step S103, the insurance year sequence refers to a set of data formed by arranging the insurance year values ​​as elements in a predetermined order. For example, if the insurance years of vehicle C are 2012, 2013, 2016 and 2018, then the insurance year sequence can be {2012, 2013, 2016, 2018} or {2018, 2016, 2013, 2012}. Specifically, in this embodiment, the insurance year sequence is obtained by sorting the insurance year values ​​from large to small or from small to large using an array sort method.

[0137] Since the user may not be insured continuously, for example, the insurance years are 2012, 2013, 2015, and 2016; although the user was not insured in 2014, resulting in only two years of continuous insurance in the usual sense, it is obvious that the user's continuous insurance probability is relatively high and should not be calculated based on two years. Therefore, a comprehensive calculation can be made based on the years of his multiple continuous insurance coverage to obtain a comprehensive continuous insurance year; or the user's insurance years are 2012, 2014, 2016, and 2018, then an equivalent continuous insurance year can also be calculated.

[0138] As described in the above steps S104 and S105, in this embodiment, the difference between adjacent insurance years in the insurance year sequence is calculated in sequence. If the difference is 1, it means that the two adjacent years are consecutive insurance years; if the difference is greater than 1, it means that there is an uninsured year between the two adjacent years, but the user still has insurance behavior in the two adjacent years. At this time, the corresponding values ​​in the difference sequence are input into the preset formula to calculate the user's consecutive insurance years, which is an equivalent consecutive insurance years. Obviously, the larger the difference, the lower the equivalent consecutive insurance years will be.

[0139] In one embodiment, the difference sequence composed of the difference values ​​is {a 1 ,a 2 ,a 3 ,…a n},

[0140] The preset formula is Formula 1: In this embodiment, the above differences can be arranged from small to large, and the arranged differences are sequentially combined into the above difference sequence {a 1 ,a 2 ,a 3 ,…a n}; This is because the smaller the difference, the greater its impact on the probability of continuous coverage, and the more it can represent the possibility of user coverage. Therefore, the smaller the difference, the greater its impact on the final calculation result during calculation. If the difference is larger, it means that the number of consecutive years of non-insurance is more, the smaller its impact on the probability of continuous coverage. Therefore, the larger the difference, the smaller its impact on the final calculation result during calculation.

[0141] In another embodiment, the preset formula for calculating the number of consecutive years of insurance coverage for a user is Formula 2:

[0142]

[0143] In this embodiment, since the above insurance year sequence is arranged from large to small, it can be understood that the closer the year is to the current time (such as 2018, 2017), the more likely it is that the user will be insured, and the user's insurance year data is more referenceable; while the insurance year farther from the current time is relatively less referenceable. Therefore, in the calculation process of the above formula, the further back (that is, the farther the insurance year is from the current year), the smaller its impact on the calculation result of the continuous insurance years of the end user will be.

[0144] The above insurance year sequence takes {2018, 2017, 2015, 2010} as an example. In the conventional definition of continuous insurance years, the above continuous insurance years are 2, because only 2018 and 2017 are continuous insurance years. In this embodiment, the difference between adjacent insurance years in the insurance year sequence is calculated as 1, 2, 5, and the difference is sequentially formed into a difference sequence of {1, 2, 5}; then the values ​​in the difference sequence are correspondingly input into Formula 2, and the continuous insurance years are calculated as: The calculated number of consecutive insurance years is obviously greater than the usual number of consecutive insurance years (ie 2), and the calculation result will obviously more truly reflect the user's actual insurance needs.

[0145] In another embodiment, after the step S105 of inputting the corresponding values ​​in the difference sequence into a preset formula to calculate the number of consecutive insurance years of the user, the method includes:

[0146] Step S107, obtaining the number of years of insurance for the user according to the insurance year of the user;

[0147] Step S108, obtaining the year span of the user's insurance according to the maximum year and the minimum year among the user's insurance years;

[0148] Step S109, calculating the user's continuous insurance probability according to the number of years the user has been insured, the number of consecutive insurance years and the year span;

[0149] Wherein, the number of years the user is insured is p, the number of consecutive years of insurance is q, the year span is m, the equivalent consecutive insurance probability is W, and the calculation formula of the consecutive insurance probability is:

[0150] It is understandable that the user's continuous coverage probability in the current year is closely related to the number of years the user was previously insured, the span of insured years, and the number of continuous insurance years. Therefore, an equivalent continuous coverage probability of the user in the current year can be calculated using the above parameters.

[0151] Taking the insurance year sequence {2018, 2017, 2015, 2010} in the above embodiment as an example, the above calculation process shows that the number of consecutive insurance years q is 73 / 24, the number of years p insured by the user is 4, and the year span m is 9. The above equivalent continuous insurance probability is That is, the current equivalent continuous underwriting probability of the user is 15%. Furthermore, according to the corresponding relationship between the equivalent continuous underwriting probability and the sales terminal, the sales terminal corresponding to the above equivalent continuous underwriting probability is determined; and the user information of the user and the terminal information of the sales terminal are obtained, and the user information of the user and the terminal information of the sales terminal are sequentially synthesized into the preset layers to generate a synthetic picture, and the synthetic picture is sent to the sales terminal and stored in the preset database. The above information is generated into a picture, so that the above information can be solidified in the same picture, which is convenient for viewing and avoiding missing or losing information; it is stored in the database for subsequent tracking and query.

[0152] In this embodiment, the classification result includes an underwriting label and a corresponding prediction probability. After the classification layer classifies and calculates the fused feature vector to obtain the classification result corresponding to the user text information as the recognition result of the voice information returned by the target user in step S7, the following steps are included:

[0153] Step S8a, weighted calculation is performed based on the continuous underwriting probability and the predicted probability corresponding to the underwriting tag to obtain the underwriting probability of the user. In this embodiment, the continuous underwriting probability calculated by the above method and the predicted probability predicted by the above recognition model are combined for comprehensive calculation, so that the final result is more referenceable and can better reflect the user's real underwriting needs.

[0154] Reference Figure 2 In one embodiment of the present application, an information identification device is also provided, including:

[0155] The selection unit 10 is used to read the continuous insurance years of each user stored in the preset database, and select the target user whose continuous insurance years are greater than the preset value;

[0156] The output unit 20 is used to output the preset sales text information to the user terminal where the target user is located;

[0157] The conversion unit 30 is used to receive the voice information returned by the target user through the user terminal, and perform voice conversion on the voice information to obtain corresponding user text information;

[0158] An input unit 40 is used to input the user text information into a preset recognition model; the recognition model is obtained by training a fusion CNN model and a DBM model, wherein a classification layer connected to the fully connected layer of the CNN model and the feature output layer of the DBM model is used as the final output layer of the recognition model;

[0159] An extraction unit 50, configured to extract a local feature vector of the user text information through the fully connected layer of the CNN model, and extract a global feature vector of the user text information through the DBM model;

[0160] A fusion unit 60, used for fusing the local feature vector with the global feature vector to obtain a fused feature vector and input it into the classification layer;

[0161] The classification unit 70 is used to perform classification calculation on the fused feature vector through the classification layer to obtain a classification result corresponding to the user text information as a recognition result of the voice information returned by the target user.

[0162] In one embodiment, the information identification device further includes:

[0163] Create a unit to create a blank layer;

[0164] A first acquisition unit, used to acquire the number of characters of the sales text information and the user text information;

[0165] A first calculation unit is used to calculate the first area size and the second area size occupied by the sales text information and the user text information when added to the blank layer according to the number of characters of the sales text information and the user text information and a preset font format;

[0166] A division unit, configured to divide the blank layer into a first area corresponding to the size of the first area and a second area corresponding to the size of the second area; and to divide a third area of ​​a preset size between the first area and the second area in the blank layer;

[0167] The adding unit is used to add the sales text information into the first area, add the user text information into the second area, add the classification result into the third area, and synthesize them into a picture.

[0168] In one embodiment, the adding unit comprises:

[0169] A calculation subunit, used to perform hash calculations on the sales text information, the user text information and the classification result, respectively, to obtain corresponding sales hash values, user hash values ​​and classification hash values;

[0170] A first adding subunit, configured to encrypt the sales text information by using the classification hash value, and add the encrypted sales text information into the first area;

[0171] a second adding subunit, configured to encrypt the user text information by using the sales hash value, and add the encrypted user text information into the second area;

[0172] The third adding subunit is used to encrypt the classification result by using the user hash value, and add the encrypted classification result to the third area.

[0173] In one embodiment, the classification result includes an underwriting label and a corresponding predicted probability; the information identification device further includes:

[0174] A storage unit, used for storing the sales text information and the user text information in the preset database;

[0175] A first combining unit, used to combine the sales text information with the underwriting label to obtain a first combination;

[0176] A second combining unit, configured to combine the user text information with the predicted probability corresponding to the underwriting label to obtain a second combination;

[0177] A second calculation unit, configured to calculate the first combination and the second combination respectively by using a hash algorithm to obtain a corresponding first hash value and a second hash value;

[0178] An uploading unit is used to add the first hash value and the second hash value into the same array to generate a first collection; and upload the first collection to the enterprise underwriting management server for storage.

[0179] In one embodiment, the information identification device further includes:

[0180] The second acquisition unit is used to acquire the user's historical insurance policy underwriting data based on the spark data computing engine;

[0181] A third calculation unit is used to calculate the user's historical insurance policy underwriting data to obtain the user's underwriting year;

[0182] A sorting unit, used for sorting the insurance years based on an array sort algorithm to obtain an insurance year sequence arranged from largest to smallest;

[0183] A fourth calculation unit is used to sequentially calculate the differences between adjacent underwriting years in the underwriting year sequence, and sequentially form the differences into a difference sequence;

[0184] A fifth calculation unit, used for inputting the corresponding values ​​in the difference sequence into a preset formula to calculate the continuous insurance years of the user;

[0185] The storage unit is used to associate the user's continuous insurance years with the user's user information and store them in a preset database.

[0186] In one embodiment, the information identification device further includes:

[0187] A third acquisition unit, configured to acquire the number of years of insurance coverage for the user according to the insurance year of the user;

[0188] A fourth acquisition unit, configured to acquire the year span of the user's insurance according to the maximum year and the minimum year of the user's insurance years;

[0189] a sixth calculation unit, configured to calculate the continuous insurance probability of the user according to the number of years of insurance of the user, the number of consecutive insurance years and the year span;

[0190] The classification result includes an underwriting label and a corresponding predicted probability, and the information identification device further includes:

[0191] The seventh calculation unit is used to perform weighted calculation according to the continuous underwriting probability and the predicted probability corresponding to the underwriting label to obtain the underwriting probability of the user.

[0192] In this embodiment, the specific implementation of the above-mentioned units / sub-units please refer to the corresponding description in the above-mentioned method embodiment, which will not be repeated here.

[0193] Reference Figure 3 In an embodiment of the present application, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store text information, etc. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the information identification method in the above method embodiment is implemented, and it will not be repeated here.

[0194] Those skilled in the art will understand that Figure 3 The structure shown in is merely a block diagram of a portion of the structure related to the present application solution and does not constitute a limitation on the computer device to which the present application solution is applied.

[0195] An embodiment of the present application further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the information identification method in the above method embodiment is implemented, which will not be repeated here. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0196] In summary, the information recognition method, device, computer equipment and storage medium provided in the embodiments of the present application receive the voice information returned by the target user, and perform voice conversion on the voice information to obtain the corresponding user text information; the user text information is respectively input into the preset recognition model; the recognition model is obtained by training the fusion CNN model and the DBM model; the local feature vector of the user text information is extracted by the fully connected layer of the CNN model, and the global feature vector of the user text information is extracted by the DBM model; the local feature vector is then fused with the global feature vector to obtain a fused feature vector and input it into the classification layer; finally, the fused feature vector is classified and calculated by the classification layer to obtain the classification result corresponding to the user text information as the recognition result of the voice information returned by the target user. The present application not only takes into account the local features in the user text information, but also combines the global features, so that the final classification result is more accurate.

[0197] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided in this application and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0198] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0199] The above description is only a preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. An information identification method, characterized in that: The following steps are involved: Read the continuous insurance years of each user stored in the preset database, and select the target user whose continuous insurance years are greater than the preset value; Outputting preset sales text information to the user terminal where the target user is located; Receiving the voice information returned by the target user through the user terminal, and performing voice conversion on the voice information to obtain corresponding user text information; Inputting the user text information into a preset recognition model respectively; The recognition model is obtained by training a fusion CNN model and a DBM model, wherein a classification layer connected to the fully connected layer of the CNN model and the feature output layer of the DBM model serves as the final output layer of the recognition model; Extracting a local feature vector of the user text information by the fully connected layer of the CNN model, and extracting a global feature vector of the user text information by the DBM model; Fusing the local feature vector with the global feature vector to obtain a fused feature vector and inputting the fused feature vector into the classification layer; The fused feature vector is classified and calculated by the classification layer to obtain a classification result corresponding to the user text information as a recognition result of the voice information returned by the target user; After the step of performing classification calculation on the fused feature vector through the classification layer to obtain the classification result corresponding to the user text information as the recognition result of the voice information returned by the target user, the method further comprises: Create a blank layer; Obtaining the number of characters of the sales text information and the user text information; Calculate the first area size and the second area size occupied by the sales text information and the user text information when added to the blank layer according to the number of characters of the sales text information and the user text information and the preset font format; Dividing the blank layer into a first area corresponding to the size of the first area and a second area corresponding to the size of the second area; and dividing a third area of ​​a preset size between the first area and the second area in the blank layer; The sales text information is added into the first area, the user text information is added into the second area, the classification result is added into the third area, and they are combined into a picture.

2. The information identification method according to claim 1, characterized in that: The step of adding the sales text information into the first area, adding the user text information into the second area, and adding the classification result into the third area comprises: Performing hash calculations on the sales text information, the user text information, and the classification result to obtain corresponding sales hash values, user hash values, and classification hash values; Encrypting the sales text information by using the classification hash value, and adding the encrypted sales text information into the first area; Encrypting the user text information by using the sales hash value, and adding the encrypted user text information into the second area; The classification result is encrypted by the user hash value, and the encrypted classification result is added to the third area.

3. The information identification method according to claim 1, characterized in that: The classification result includes an insurance label and a corresponding prediction probability; after the step of performing classification calculation on the fused feature vector through the classification layer to obtain the classification result corresponding to the user text information as the recognition result of the voice information returned by the target user, the method includes: Saving the sales text information and user text information in the preset database; Combining the sales text information with the underwriting label to obtain a first combination; Combining the user text information with the predicted probability corresponding to the underwriting label to obtain a second combination; Calculate the first combination and the second combination respectively by using a hash algorithm to obtain a corresponding first hash value and a second hash value; The first hash value and the second hash value are added to the same array to generate a first collection; and the first collection is uploaded to the enterprise underwriting management server for storage.

4. The information identification method according to claim 1, characterized in that: Before the step of reading the continuous insurance years of each user stored in the preset database and selecting the target user whose continuous insurance years are greater than the preset value, the method includes: Based on the spark data computing engine, obtain the user's historical insurance policy underwriting data; Calculate the user's historical insurance policy underwriting data to obtain the user's underwriting year; Based on the array sort algorithm, the insurance years are sorted to obtain an insurance year sequence arranged from largest to smallest; Calculating the differences between adjacent underwriting years in the underwriting year sequence in sequence, and sequentially composing the differences into a difference sequence; Inputting the corresponding values ​​in the difference sequence into a preset formula to calculate the number of consecutive years of insurance coverage for the user; The user's continuous insurance years are associated with the user's user information and stored in a preset database.

5. The information identification method according to claim 4, characterized in that: After the step of inputting the corresponding values ​​in the difference sequence into a preset formula to calculate the number of consecutive years of insurance coverage for the user, the method further comprises: According to the insurance year of the user, obtain the number of years the user is insured; According to the maximum year and the minimum year among the insured years of the user, obtain the year span of the user's insurance; Calculate the user's continuous insurance probability according to the number of years the user has been insured, the number of consecutive years of insurance, and the year span; The classification result includes an underwriting label and a corresponding prediction probability. After the step of performing classification calculation on the fused feature vector through the classification layer to obtain the classification result corresponding to the user text information as the recognition result of the voice information returned by the target user, the method includes: The underwriting probability of the user is obtained by performing a weighted calculation based on the continuous underwriting probability and the predicted probability corresponding to the underwriting label.

6. An information identification device, used to implement the method according to any one of claims 1 to 5, characterized in that: include: A selection unit is used to read the continuous insurance years of each user stored in the preset database, and select a target user whose continuous insurance years are greater than a preset value; An output unit, used for outputting preset sales text information to a user terminal where the target user is located; A conversion unit, configured to receive the voice information returned by the target user through the user terminal, and perform voice conversion on the voice information to obtain corresponding user text information; An input unit, used to input the user text information into a preset recognition model respectively; The recognition model is obtained by training a fusion CNN model and a DBM model, wherein a classification layer connected to the fully connected layer of the CNN model and the feature output layer of the DBM model serves as the final output layer of the recognition model; An extraction unit, configured to extract a local feature vector of the user text information through the fully connected layer of the CNN model, and extract a global feature vector of the user text information through the DBM model; A fusion unit, used for fusing the local feature vector with the global feature vector to obtain a fused feature vector and input it into the classification layer; The classification unit is used to perform classification calculation on the fused feature vector through the classification layer to obtain a classification result corresponding to the user text information as a recognition result of the voice information returned by the target user.

7. The information recognition device according to claim 6, characterized in that: Also includes: Create a unit to create a blank layer; A first acquisition unit, used to acquire the number of characters of the sales text information and the user text information; A first calculation unit is used to calculate the first area size and the second area size occupied by the sales text information and the user text information when added to the blank layer according to the number of characters of the sales text information and the user text information and a preset font format; a dividing unit, configured to divide the blank layer into a first area corresponding to the first area size and a second area corresponding to the second area size; and dividing a third area of ​​a preset size between the first area and the second area in the blank layer; The adding unit is used to add the sales text information into the first area, add the user text information into the second area, add the classification result into the third area, and synthesize them into a picture.

8. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.