Information processing method, device and electronic equipment
By converting the streaming program sequences in the intelligent device into a specific information format and using a neural network model for evaluation, the problems of low efficiency and low accuracy of streaming program sequence evaluation in the existing technology are solved, and automatic rating and operation of the intelligent customer service process are realized Improved dimensional efficiency.
Patent Information
- Application Number
- CN201911337129.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2019-12-23
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2040-08-10
AI Technical Summary
The prior art cannot effectively evaluate the streaming program sequences added to smart devices, and due to different writing languages, it is impossible to use a unified method to evaluate and process, resulting in low efficiency and low accuracy.
By obtaining the information composed of nodes with associated relationships, converting them into target information in a specific information format, and using the neural network model obtained through sample information training for evaluation and processing, the evaluation information is obtained. The sample information obtained by the conversion of log data has a specific correspondence with the nodes of the original information, and realizes effective evaluation and format conversion of the composition information of different nodes.
It improves the efficiency and accuracy of information evaluation by processing equipment, can automatically rating intelligent customer service processes, and improves the operational efficiency and user experience of operation and maintenance personnel.
Smart Images

Figure CN111046162B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of information processing technology, and more specifically to an information processing method, device and electronic equipment. Background Art
[0002] With the development of information technology, smart devices can increasingly assist users' lives and learning. Therefore, operation and maintenance personnel will add more and more process sequences to smart devices so that they can generate more smart applications.
[0003] However, the existing technology cannot effectively evaluate the added process sequences, and relies more on manual evaluation. In addition, since different process sequences correspond to different writing languages, it is impossible to use a unified method to evaluate and process them, which makes the efficiency of evaluating process sequences using processing equipment in the existing technology low and the accuracy is not high. Summary of the invention
[0004] In view of this, this application provides the following technical solutions:
[0005] An information processing method, comprising:
[0006] Acquire first information, where the first information consists of nodes with an associated relationship;
[0007] converting the first information into target information according to a conversion condition of a specific information format;
[0008] Processing the target information using a pre-created evaluation model to obtain evaluation information corresponding to the first information, wherein the evaluation model is a neural network model obtained by training sample information, the sample information is obtained by converting log data, and the nodes of the log data have a specific corresponding relationship with the nodes of the first information;
[0009] The evaluation information is output.
[0010] Optionally, the method further includes training the evaluation model, including:
[0011] Convert the acquired log data into sample information according to the conversion conditions of the specific information format;
[0012] Determine evaluation information corresponding to each sample information in the sample information;
[0013] Based on the sample information and the evaluation information corresponding to each sample information, a neural network model is trained to obtain an evaluation model.
[0014] Optionally, the method further comprises:
[0015] Perform node splitting on the first information according to a specific splitting condition to obtain split nodes, wherein the characteristic splitting condition indicates that each node after the splitting can realize the function of part of the first information;
[0016] Type statistics are performed on the disassembled nodes, and nodes of the same type are classified to obtain multiple nodes of the first information.
[0017] Optionally, the converting the first information into target information according to the conversion condition of the specific information format includes:
[0018] Matching a node label corresponding to a specific information format for each node of the first information;
[0019] Generate target information according to the node label.
[0020] Optionally, the training a neural network model based on the sample information and the evaluation information corresponding to each sample information to obtain an evaluation model includes:
[0021] Determine the node label of each sample information;
[0022] Determine evaluation information corresponding to each node label of the sample information;
[0023] Based on the sample information, node labels and the evaluation information corresponding to each node label, a neural network model is trained to obtain an evaluation model.
[0024] Optionally, the method further comprises:
[0025] Performing node hierarchical evaluation on the first information using a pre-created evaluation model to obtain a target node, where the target node is a key node in the first information determined according to the pre-created evaluation model;
[0026] The target node is outputted so that information matching the target node can be obtained to update the first information.
[0027] Optionally, converting the first information into target information in a specific information format includes:
[0028] The first information is processed using a pre-created processing model to obtain target information matching the first information, wherein the processing model is a neural network model obtained by training format sample information, and each sample information of the format sample information includes second information and target information in a specific format marked to correspond to the second information.
[0029] Optionally, the specific format includes an information description format, and the method further includes:
[0030] Get information in information description format;
[0031] The information in the information description format is input into the pre-created processing model to obtain third information matching the information in the information description format.
[0032] An information processing device, comprising:
[0033] An information acquisition unit, configured to acquire first information, wherein the first information is composed of nodes having an associated relationship;
[0034] a conversion unit, configured to convert the first information into target information according to a conversion condition of a specific information format;
[0035] a processing unit, configured to process the target information using a pre-created evaluation model to obtain evaluation information corresponding to the first information, wherein the evaluation model is a neural network model obtained by training sample information, the sample information is obtained by converting log data, and the nodes of the log data have a specific corresponding relationship with the nodes of the first information;
[0036] An output unit is used to output the evaluation information.
[0037] An electronic device, comprising:
[0038] Memory, used to store programs;
[0039] A processor is used to execute the program, wherein the program is specifically used to:
[0040] Acquire first information, where the first information consists of nodes with an associated relationship;
[0041] converting the first information into target information according to a conversion condition of a specific information format;
[0042] Processing the target information using a pre-created evaluation model to obtain evaluation information corresponding to the first information, wherein the evaluation model is a neural network model obtained by training sample information, the sample information is obtained by converting log data, and the nodes of the log data have a specific corresponding relationship with the nodes of the first information;
[0043] The evaluation information is output.
[0044] Through the above technical solutions, it can be known that the present application discloses an information processing method, device and electronic device, which obtains first information composed of nodes with an associated relationship; converts the first information into target information in a specific information format, so that the evaluation model processes the target information, obtains evaluation information corresponding to the first information, and outputs the evaluation information. It is realized that the information composed of different nodes can be effectively evaluated through the evaluation model, and the information can be formatted, and it is also realized that information of different formats can be processed, which improves the processing efficiency and accuracy of the processing equipment in evaluating and processing information. BRIEF DESCRIPTION OF THE DRAWINGS
[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.
[0046] Figure 1 A schematic diagram of a scenario of human-computer interaction of an intelligent customer service is provided for an embodiment of the present application;
[0047] Figure 2 A flowchart of an information processing method provided in an embodiment of the present application;
[0048] Figure 3 A schematic diagram of a scenario of an evaluation model training process provided in an embodiment of the present application;
[0049] Figure 4 A schematic diagram of the structure of an information processing device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0050] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0051] At present, human-computer conversation can be carried out through the intelligent conversation system. For example, the intelligent conversation system is applied to electronic devices, such as installing the intelligent conversation system in the electronic device in the form of an APP (application), and when the electronic device runs the APP, the user can conduct a human-computer conversation through the electronic device. Correspondingly, the intelligent conversation system can also be a conversation process in a business APP installed in the electronic device. For example, there will be a conversation process for communicating with intelligent customer service in a shopping APP. More and more intelligent conversation systems, users and clients use intelligent customer service to solve problems. In order to make the intelligent customer service process meet the needs of the scene, the intelligent customer service operation and maintenance personnel will add or modify the intelligent customer service process. The evaluation information of the intelligent customer service process can prompt the intelligent customer service operation and maintenance personnel to better maintain the intelligent customer service process. In the embodiment of the present application, automatic rating of the intelligent customer service process is realized, which can improve the operation and maintenance efficiency of the intelligent customer service process by the operator and improve the user experience.
[0052] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings of the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0053] See also Figure 1 , which shows a schematic diagram of a human-computer interaction scenario of an intelligent customer service provided in an embodiment of the present application. In this scenario 100, a user can perform human-computer interaction with the intelligent customer service. Figure 1 ,The information interaction process between users and intelligent customer service is as follows:
[0054] User: Hello.
[0055] Intelligent Customer Service: Hello, how can I help you?
[0056] User: I want to find out the logistics information of my package.
[0057] Smart Customer Service: Which package is this?
[0058] User: Package number 123**456.
[0059] Intelligent customer service: The package was sent from city A on September 6, arrived at the sorting center in city B on September 7, and arrived at the destination outlet on September 8. It is being delivered.
[0060] The entire process information of the interaction between the intelligent customer service and the user is the information that needs to be processed in the embodiment of the present application. The information processing process is described in detail below.
[0061] See also Figure 2 , which shows a flow chart of an information processing method provided by an embodiment of the present application, the method may include the following steps:
[0062] S201. Obtain first information.
[0063] The first information is composed of nodes with an association relationship. The first information in the embodiment of the present application is information that can be node-divided, that is, it is information composed of several nodes. The nodes in the first information are nodes with a certain association relationship. The association relationship can represent the association relationship between each node, and can also represent the association relationship between each node and the first information. For example, when the association relationship represents the association relationship between each node, it can be the sequence or causal relationship between each node. When the association relationship represents the association relationship between each node and the first information, that is, each node can reflect the complete semantics of the first information after combination.
[0064] Taking the first information as an intelligent customer service conversation process as an example, the intelligent customer service conversation process includes several nodes, and each node realizes the function of the intelligent customer service conversation process through the connection between each other. Figure 1 The intelligent customer service conversation process shown in the figure includes three nodes, namely "greeting", "asking questions" and "returning questions". The corresponding "greeting" node presents the message "Hello" to the user; the "asking questions" node presents the message "How can I help you" to the user; the "returning questions" node presents the message "Which package is it?" to the user.
[0065] S202: Convert the first information into target information according to a conversion condition of a specific information format.
[0066] S203: Use a pre-created evaluation model to process the target information to obtain evaluation information corresponding to the first information.
[0067] The first information may be process information written by operation and maintenance personnel according to scenario requirements. Due to different formats of the writing language or different operating scenarios of the first information, the first information cannot be evaluated in a unified manner during the evaluation process and can only be evaluated manually or requires professional personnel to perform analysis and evaluation.
[0068] In the embodiment of the present application, a unified evaluation model is used when evaluating the first information. In order to facilitate the evaluation of the first information by the evaluation model, the first information needs to be converted into target information. During the conversion process, the conversion conditions of the specific information format are used, and the conversion conditions of the specific information format represent the conversion conditions of the target information. Since the target information needs to be used as input information of the evaluation model, the conversion conditions are essentially matched with the data processing format of the evaluation model.
[0069] For example, if the evaluation model implements the processing of natural language, the first information needs to be converted into natural language. Specifically, each node of the first information can be compared to a letter (or word) in a natural language, so that the entire first information can be considered as a natural language. Specifically, the processing flow of converting the first information into the target information will be described in detail in the subsequent embodiments of this application.
[0070] After converting the first information into target information, the target information is processed using a pre-created evaluation model to obtain evaluation information corresponding to the first information. The evaluation model is a neural network model obtained by training sample information. The sample information for training the neural network model is obtained by converting log data, and the nodes of the log data have a specific corresponding relationship with the nodes of the first information. The log data comes from the application system or application of the first information. If the first information represents the conversation process between the intelligent customer service and the user, the log information can come from the data loading in the intelligent customer service application and system. It should be noted that the log data includes the user's evaluation of the conversation process of the intelligent customer service. That is, taking the intelligent interactive APP as an example, the user log is analyzed by obtaining the data of the APP to find out the user's evaluation of the process. In addition, the nodes of the log data have a specific corresponding relationship with the nodes of the first information, and the corresponding relationship can be that the nodes of the log data and the nodes of the first information use the same processing method when dividing the nodes, or the nodes of the log data and the nodes of the first information have the same classification conditions, or the nodes of the log data and the nodes of the first information have a relationship of inclusion or being included, or the nodes of the log data and the nodes of the first information come from the same node database. This indicates that the source data log data of the sample information of the evaluation model has the same attributes as the first information, that is, both can be converted into the format of the target information and are composed of nodes with an associated relationship.
[0071] Then the log data is converted into the information format of the target information. The log data is used as sample information. When training the neural network, the sample information can be divided into a training set and a test set. Then the neural network is trained to obtain an evaluation model. The evaluation model can be modeled using classification algorithms such as decision trees, logistic regression, Portsmouth Bayes, neural networks, support vector machines SVM (Support Vector Machine), K-nearest neighbor KNN (k-NearestNeighbor) and other algorithms. For example, these sample information are used as input sample data of the classification algorithm SVM or KNN, and model training is performed to construct a classifier, that is, an evaluation model, and the first information collected above that needs to be evaluated is input into the evaluation model for processing, thereby obtaining evaluation information for the first information.
[0072] S204: Output evaluation information.
[0073] The evaluation information can be matched with the first information and output. In the intelligent customer service application scenario, each conversation process can be evaluated, so as to assist the operation and maintenance personnel to reconfirm or improve the added or modified conversation process according to the evaluation results.
[0074] It should be noted that each time the operation and maintenance personnel generates a session process, the process can be evaluated through the evaluation model, and then the evaluation information can be displayed at the location corresponding to the process, so that the operation and maintenance personnel can obtain the evaluation information of the generated process in real time. The evaluation information can also be output during the test of the generated process. The evaluation information can be output to the current electronic device or to a specified electronic device, that is, as long as the corresponding output path is set for the evaluation information.
[0075] The embodiment of the present application discloses an information processing method, which obtains first information composed of nodes with an associated relationship; converts the first information into target information in a specific information format, so that the evaluation model processes the target information, obtains evaluation information corresponding to the first information, and outputs the evaluation information. It is realized that the information composed of different nodes can be effectively evaluated through the evaluation model, and the information can be formatted, and it is also realized that information of different formats can be processed, which improves the processing efficiency and accuracy of the processing equipment in evaluating and processing information.
[0076] For ease of description, in the following description of each embodiment, the first information is described as the intelligent customer service conversation process in the intelligent customer service application scenario. Figure 3 , which shows a schematic diagram of a scenario of the evaluation model training process in an embodiment of the present application, the training process includes:
[0077] Log data 302 is obtained from server 301. The log data 302 may include historical user evaluation information on the intelligent customer service conversation process, and may also include target user or target user group evaluation information on a specific intelligent customer service conversation process, that is, the log data 302 includes not only the intelligent customer service conversation process but also the corresponding evaluation information. Specific users or specific conversation processes may be selected as training samples according to the needs of the application scenario, so that the trained model can be applied to the evaluation of the entire intelligent customer service, certain target users, and specific conversation processes.
[0078] After obtaining the log data 302, the log data 302 is converted into sample information 303 according to the conversion conditions of the specific information format, wherein the conversion conditions of the specific information format match the information format or information processing mode of the converted sample information. Then the sample information is matched with the evaluation information. The sample information and the evaluation information corresponding to each sample information are used as the input of the model, that is, the sample information marked with the evaluation information is used as the input of the training model, for example, input into the convolutional neural network 304 (Convolutional Neural Networks, CNN) for training to obtain the evaluation model 305.
[0079] It should be noted that the process can apply machine learning methods, that is, use and its learning method to establish a classifier corresponding to the first information. The function of the classifier is to extract features from the first information, and then classify the features into corresponding evaluation information, thereby obtaining evaluation information corresponding to the first information. The machine learning methods include: artificial neural network method, support vector machine method, decision tree method, Bayesian method, random forest method, and logistic regression method. In the specific implementation process, different machine learning methods can also be used to establish classifiers corresponding to the same sample information; for example, a classifier of an artificial neural network method can be established, a classifier of a Bayesian method can be established, and a classifier of a random forest method can be established; then the classifier with the highest accuracy rate is used as the classifier corresponding to the first information.
[0080] Since the first information is composed of nodes with an associated relationship in the embodiment of the present application, the first information can also be node-splitting to obtain its constituent nodes. In actual application, if the first information needs to be debugged and modified, the corresponding node information can be modified without modifying the entire information.
[0081] Specifically, an embodiment of the present application further provides a method for information splitting, which may include the following steps:
[0082] S401, splitting the first information into nodes according to a specific splitting condition to obtain split nodes;
[0083] S402: Perform type statistics on the split nodes, classify the nodes of the same type, and obtain multiple nodes of the first information.
[0084] The specific splitting condition represents that each node after the splitting can realize the information function of the first information part.
[0085] For example, the first information represents the intelligent customer service conversation process, and the intelligent customer service conversation process can be split into nodes, and the process corresponding to each sub-function can be determined as a node. For example, a node representing greeting information, a node representing question information, a node representing counter-question information, a node representing request evaluation, and so on. In order to facilitate the analysis of information, it is necessary to count the split nodes according to type, and of course, it can also be counted according to other statistical conditions, such as according to the modification time of the node. In this way, the nodes can be classified, which is convenient for subsequent application and analysis.
[0086] Specifically, the converting of the first information into the target information according to the conversion condition of the specific information format includes:
[0087] S501, matching a node label corresponding to a specific information format for each node of the first information;
[0088] S502: Generate target information based on the node label.
[0089] After obtaining the first information, the first information is split to obtain the various nodes of the first information, and then the nodes are counted by type, and a label matching the format of the target information is set for each type of node. For example, the target information represents the format of natural language. Then a number can be set for each type of node of the first information, and these numbers are equivalent to letters (or words) in natural language, so the first information can be converted into a format similar to an article in natural language. In this way, the process nodes can be regarded as words, and the entire process can be regarded as a sentence, which can be used to evaluate the quality of the process based on natural language processing.
[0090] In some embodiments, the training of a neural network model to obtain an evaluation model based on the sample information and the evaluation information corresponding to each sample information includes:
[0091] Determine the node label of each sample information;
[0092] Determine the evaluation information corresponding to the node label of each sample information;
[0093] Based on the sample information, node labels and the evaluation information corresponding to each node label, the neural network model is trained to obtain the evaluation model.
[0094] For example, the process nodes of intelligent customer service can be represented as a new language, and the intelligent customer service process can be automatically rated according to the automatic grading of articles, which makes it convenient to evaluate and assess the processes added by intelligent customer service operators, improve the enthusiasm of intelligent customer service operators, and improve the satisfaction of intelligent customer service.
[0095] Specifically, by analyzing user logs, we can find out the users' evaluations of the intelligent customer service process. Then, we can count the process nodes by type, set a label for each type, and compare these labels to letters in a language. In this way, we can represent a process as an article, and then set a level for the process. We can train it according to the idea of article grading, and the resulting evaluation model can output the evaluation information of each process and the nodes in each process.
[0096] In order to facilitate the operation and maintenance personnel to control the intelligent customer service process, in the embodiment of the present application, the process can also be evaluated by node classification. The specific process includes:
[0097] S601, using a pre-created evaluation model to perform node classification evaluation on the first information to obtain a target node;
[0098] S602: Output the target node so as to obtain information matching the target node to update the first information.
[0099] The target node is a key node of the first information determined according to a pre-created evaluation model. Since the evaluation model can obtain evaluation information for each node according to the process information corresponding to each node in the first information, if the evaluation information meets a certain condition, the node is characterized as a key node in the first information. For example, the evaluation information represents the user's satisfaction with the node. If the evaluation information obtained for a certain node represents that the user's satisfaction with the node is poor, the node is determined as the target node, that is, the key node of the first information. Then, after obtaining the target node, the operation and maintenance personnel can change the process information corresponding to the node, that is, obtain the matching information with the target node, so as to improve the entire first information, that is, improve the entire intelligent customer service interaction process.
[0100] For example, the operation and maintenance personnel designed a set of intelligent customer service interaction processes, which include greeting nodes, question nodes, and counter-question nodes. The intelligent customer service interaction process is input into the evaluation model. If the score of the greeting node is low, the operation and maintenance personnel can know that the specific process information corresponding to the node does not meet the actual needs of users, or makes the user experience poor, and then modify the process information corresponding to the node. For example, the original process information corresponding to the greeting node is that the user sends a greeting message, and the intelligent customer service directly generates a question. The corresponding modification can be to first feedback the greeting message after receiving the user's greeting message, and then generate a question.
[0101] Correspondingly, when converting the first information into target information of a specific information format, the first information can also be processed using a pre-created processing model to obtain target information that matches the first information. The processing model is a neural network model obtained by training the format sample information. Each sample information of the format sample information and the target information of a specific format corresponding to the second information are annotated. That is, by learning the features between the original information format and the target information format through a machine learning method, the format conversion process can be automatically completed through the neural network model. The creation process of the processing model can be used as a reference based on the creation process of the evaluation model, and no tracing is performed in this embodiment.
[0102] It should be noted that when creating a processing model, the format of the target information to be learned can be selected based on actual needs, that is, the information format in the sample directly affects the conversion process of the format information that can be completed by the processing model.
[0103] In some embodiments, the specific format includes an information description format, and the method further includes:
[0104] Get information in information description format;
[0105] The information in the information description format is input into a pre-created processing model to obtain third information matching the information in the information description format.
[0106] In the intelligent customer service application scenario, if a complete conversation process cannot be obtained, a description of the conversation process can be generated according to the needs, and then input into the pre-created processing model. According to the description information, the corresponding third information can be obtained, that is, the conversation process matching the description information. For example, the information in the information description format is "Conversation for user's logistics information query", and the description information is input into the processing model, and the third information obtained is the process sequence {greeting, asking, asking, querying}.
[0107] The information processing method provided in the embodiment of the present application is described below with reference to a specific application scenario.
[0108] In one of the application scenarios, the evaluation of the intelligent customer service process is realized. In this scenario, firstly, the process of users evaluating the intelligent customer service in the log of the intelligent customer service is counted, and then the evaluation information of users of each process is counted, such as the number of users' favorable comments. This is just an exemplary description, including but not limited to users' favorable comments, and other forms of evaluation methods can also be included. Then, the quality of the process is evaluated based on the user's evaluation information. After obtaining these statistical data, the level can be divided based on experts and relevant personnel. Each process can be set to a level. The process node is analogous to a letter, and the process sequence can be considered as a language. Then the rating of the process can be regarded as a label, and the sample information of <process sequence, label> can be constructed. Then, the evaluation model can be constructed according to the relevant algorithm of article grading, and the process can be graded, that is, the evaluation information is obtained. Among them, the relevant algorithm of article grading includes but is not limited to classification algorithm, machine learning, deep learning and the combination of the above algorithms.
[0109] For example, after the intelligent customer service operator has completed adding the process sequence of <“Say hello”, “Ask questions”, “Ask the model”, “Produce answers”>, the process sequence can be converted into a method that can be recognized by the evaluation model, such as using a feature vector. After calculation, the model outputs the evaluation information or evaluation level of this process sequence.
[0110] In this scenario, the process nodes of intelligent customer service are regarded as a new language. The intelligent customer service process is automatically rated according to the automatic grading of articles, which facilitates the evaluation and assessment of the processes added by intelligent customer service operators, improves the enthusiasm of intelligent customer service operators, and improves the satisfaction of intelligent customer service.
[0111] In another application scenario, the key process nodes of the added process sequence can also be extracted according to the information processing method provided in the embodiment of the present application. Specifically, by analyzing the user log, all intelligent customer service conversation processes are found, and the process nodes are counted by type, and a number is set for each type. These numbers are represented as letters in the language, so that a process sequence can be represented as a sentence, and the model is trained by extracting keywords to extract keywords, that is, the key nodes of the process. After adding the complete process sequence, the intelligent customer service operator can extract the key process nodes.
[0112] Specifically, when the process sequence is represented as a sentence, a corpus including all "sentences" can be generated, and the keywords can be extracted using the relevant algorithm for extracting keywords in natural language processing. For example, the TF-IDF method is used. TF-IDF is a statistical method used to evaluate the importance of a word to a document set or one of the documents in a corpus. The importance of a word increases in direct proportion to the number of times it appears in a document, but at the same time it decreases in inverse proportion to the frequency of its appearance in the corpus. Various forms of TF-IDF weighting are often used by search engines as a measure or rating of the relevance between a document and a user query. In addition to TF-IDF, search engines on the Internet also use a rating method based on link analysis to determine the order in which documents appear in search results. Specifically, the process of the TF-IDF algorithm is: calculate the word frequency, corresponding to this application scenario, the number of occurrences of the process node in the process sequence is calculated, then the inverse document frequency is calculated, and then the product of the word frequency and the inverse document frequency is calculated to obtain TF-IDF, thereby extracting the key nodes in the process sequence.
[0113] For example, after the intelligent customer service operator adds a "connect to wifi" process sequence, the key node model of the intelligent customer service process will first convert the relevant nodes in this process into a form that can be recognized by the TF-IDF model, and then extract the key nodes when passing them into the model. Here, the "ask about the model" node is extracted as the key node, which meets the expectations of the intelligent customer service operator. This result is informed to the tester to facilitate the tester to prepare relevant test data.
[0114] By treating the process nodes of intelligent customer service as a new language and following the relevant algorithms for extracting keywords in natural language, after the intelligent customer service operator adds a process sequence, the key process nodes in the process sequence are extracted, which is convenient for the intelligent customer service operator to view and for relevant testers to test the newly added process sequence, thereby improving the satisfaction of the intelligent customer service.
[0115] In another application scenario, the information processing method provided in the embodiment of the present application can also be used to implement a process in which intelligent customer service operation and maintenance personnel automatically generate relevant process sequences through natural language descriptions. The process includes:
[0116] By analyzing user logs, we can find out the description of relevant processes by intelligent customer service operators and the corresponding complete processes; count the process nodes by type (or other forms), and assign a number to each type; compare these numbers to letters (or words) in a language, which are represented as the target language, and the description of the process sequence by the intelligent customer service operator is represented as the source language, and train according to the relevant sequence-to-sequence algorithm to obtain a model; after the intelligent customer service operator completes the natural language description, the description information is input into the model to generate the relevant process sequence.
[0117] Count the process nodes by type (or other forms) and assign a number to each type. By analogizing these numbers to letters (or words) in a language, we can think of the process sequence as a natural language, and a process sequence is a "sentence". This "sentence" and its related description are a "sentence pair", that is, <source language, target language>. And these "sentence pairs" are the corpus. We can train in a sequence-to-sequence manner (including but not limited to RNN, Transformer, etc.). After the intelligent customer service operator completes the natural language description of the process, the intelligent customer service generation module pre-processes the natural language description and sends it to the model. The model will generate multiple process sequences in a sequence-to-sequence manner. The intelligent customer service process filtering module will exclude those that obviously do not meet the requirements in these intelligent customer service processes. If there are multiple processes, the intelligent customer service process sorting module will sort these processes and select the best ones.
[0118] For example, when the intelligent customer service operator inputs "This is a process for connecting to wifi", the sentence is first converted into a form that the model can recognize (including but not limited to the form of vectors), and then multiple process sequences are generated through sequence-to-sequence. The filtering module will filter out process sequences that do not meet the requirements. After filtering, the module will enter the sorting module for relevant sorting.
[0119] This scenario implementation method represents the process nodes of the intelligent customer service as letters in a language, so the process nodes are a new language, and the relevant processes are generated in a sequence-to-sequence manner, so that the intelligent customer service operators can automatically generate relevant processes through natural language descriptions, thereby improving the operability of the operators and improving the satisfaction of the intelligent customer service.
[0120] Correspondingly, an information processing device is also provided in the embodiment of the present application, see Figure 4 , the device comprises:
[0121] The information acquisition unit 701 is used to acquire first information, where the first information consists of nodes with an associated relationship;
[0122] A conversion unit 702, configured to convert the first information into target information according to a conversion condition of a specific information format;
[0123] The processing unit 703 is used to process the target information by using a pre-created evaluation model to obtain evaluation information corresponding to the first information, wherein the evaluation model is a neural network model obtained by training sample information, the sample information is obtained by converting log data, and the nodes of the log data have a specific corresponding relationship with the nodes of the first information;
[0124] The output unit 704 is used to output the evaluation information.
[0125] Based on the above embodiment, the device further includes a first training unit for training the evaluation model; the first training unit includes:
[0126] The conversion subunit is used to convert the acquired log data into sample information according to the conversion conditions of the specific information format;
[0127] A determination subunit, used to determine evaluation information corresponding to each sample information in the sample information;
[0128] The training subunit is used to train the neural network model to obtain an evaluation model based on the sample information and the evaluation information corresponding to each sample information.
[0129] Based on the above embodiment, the device further includes:
[0130] A node splitting unit, used for performing node splitting on the first information according to a specific splitting condition to obtain split nodes, wherein the specific splitting condition indicates that each node after the splitting can realize the function of part of the first information;
[0131] The node statistics unit is used to perform type statistics on the split nodes, classify the nodes of the same type, and obtain multiple nodes of the first information.
[0132] Based on the above embodiment, the conversion unit is specifically used for:
[0133] matching a node label corresponding to a specific information format for each node of the first information;
[0134] Generate target information according to the node label.
[0135] Based on the above embodiment, the first training subunit is specifically used for:
[0136] Determine the node label of each sample information;
[0137] Determine evaluation information corresponding to each node label of the sample information;
[0138] Based on the sample information, node labels and the evaluation information corresponding to each node label, a neural network model is trained to obtain an evaluation model.
[0139] Based on the above embodiment, the device further includes:
[0140] A hierarchical evaluation subunit, configured to perform node hierarchical evaluation on the first information using a pre-created evaluation model to obtain a target node, wherein the target node is a key node in the first information determined according to the pre-created evaluation model;
[0141] The node output subunit is used to output the target node so as to obtain information matching the target node to update the first information.
[0142] Based on the above embodiment, the conversion unit is specifically used for:
[0143] The first information is processed using a pre-created processing model to obtain target information matching the first information, wherein the processing model is a neural network model obtained by training format sample information, and each sample information of the format sample information includes second information and target information in a specific format marked to correspond to the second information.
[0144] Based on the above embodiment, the specific format includes an information description format, and the device further includes:
[0145] An acquisition subunit is used to acquire information in an information description format;
[0146] The model processing subunit is used to input the information in the information description format into the pre-created processing model to obtain third information matching the information in the information description format.
[0147] In an embodiment of the present application, an electronic device is also provided, including:
[0148] Memory, used to store programs;
[0149] A processor is used to execute the program, wherein the program is specifically used to:
[0150] Acquire first information, where the first information consists of nodes with an associated relationship;
[0151] converting the first information into target information according to a conversion condition of a specific information format;
[0152] Processing the target information using a pre-created evaluation model to obtain evaluation information corresponding to the first information, wherein the evaluation model is a neural network model obtained by training sample information, the sample information is obtained by converting log data, and the nodes of the log data have a specific corresponding relationship with the nodes of the first information;
[0153] The evaluation information is output.
[0154] The program is also used to execute the information processing method described in any of the above embodiments and the steps and methods associated with the information processing method.
[0155] In an embodiment of the present application, a storage medium is further provided, wherein the storage medium stores computer program code, and when the computer program code is executed, the information processing method as described in any one of the above is implemented.
[0156] The storage medium refers to a computer storage medium, which may contain a propagated data signal containing computer program code, such as in baseband or as part of a carrier wave. The propagated signal may have a variety of forms, including electromagnetic, optical, etc., or a suitable combination. The computer storage medium can be any computer-readable medium other than a computer-readable storage medium, which can be connected to an instruction execution system, device or equipment to communicate, propagate or transmit the program for use. The program code located on the computer storage medium can be transmitted via any suitable medium, including radio, cable, fiber optic cable, RF, or similar media, or any combination of the above media.
[0157] Each embodiment in this specification focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0158] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. An information processing method, comprising: Acquire first information, where the first information consists of nodes with an associated relationship; Converting the first information into target information according to a conversion condition of the information format; wherein the conversion condition matches a data processing format of a pre-created evaluation model; Processing the target information using a pre-created evaluation model to obtain evaluation information corresponding to the first information, wherein the evaluation model is a neural network model obtained by training sample information, the sample information is obtained by converting log data, the nodes of the log data have a corresponding relationship with the nodes of the first information, and the log data at least includes the evaluation information corresponding to the first information; outputting the evaluation information; The converting of the first information into target information according to the information format conversion condition includes: Matching a node label corresponding to the information format for each node of the first information; Generate target information according to the node label.
2. The method according to claim 1, further comprising training the evaluation model, comprising: Convert the acquired log data into sample information according to the conversion conditions of the information format; Determine evaluation information corresponding to each sample information in the sample information; Based on the sample information and the evaluation information corresponding to each sample information, a neural network model is trained to obtain an evaluation model.
3. The method according to claim 1, further comprising: Perform node splitting on the first information according to a splitting condition to obtain split nodes, wherein the splitting condition indicates that each node after the splitting can realize the function of part of the first information; Type statistics are performed on the split nodes, and nodes of the same type are classified to obtain multiple nodes of the first information.
4. The method according to claim 2, wherein the step of training a neural network model to obtain an evaluation model based on the sample information and the evaluation information corresponding to each sample information comprises: Determine the node label of each sample information; Determine evaluation information corresponding to each node label of the sample information; Based on the sample information, node labels and the evaluation information corresponding to each node label, a neural network model is trained to obtain an evaluation model.
5. The method according to claim 4, further comprising: Performing node hierarchical evaluation on the first information using a pre-created evaluation model to obtain a target node, where the target node is a key node in the first information determined according to the pre-created evaluation model; The target node is outputted so that information matching the target node can be obtained to update the first information.
6. The method according to claim 1, wherein converting the first information into target information in an information format comprises: The first information is processed using a pre-created processing model to obtain target information matching the first information, wherein the processing model is a neural network model obtained by training format sample information, and each sample information of the format sample information includes second information and target information marked with an information format corresponding to the second information.
7. The method according to claim 6, wherein the information format comprises an information description format, and the method further comprises: Get information in information description format; The information in the information description format is input into the pre-created processing model to obtain third information matching the information in the information description format.
8. An information processing device, comprising: An information acquisition unit, used to acquire first information, where the first information consists of nodes with an associated relationship; A conversion unit, configured to convert the first information into target information according to a conversion condition of the information format; wherein the conversion condition matches a data processing format of a pre-created evaluation model; a processing unit, configured to process the target information using a pre-created evaluation model to obtain evaluation information corresponding to the first information, wherein the evaluation model is a neural network model obtained by training sample information, the sample information is obtained by converting log data, the nodes of the log data correspond to the nodes of the first information, and the log data at least includes the evaluation information corresponding to the first information; An output unit, used for outputting the evaluation information; Wherein, the conversion unit is specifically used for: Matching a node label corresponding to the information format for each node of the first information; Generate target information according to the node label.
9. An electronic device, comprising: Memory, used to store programs; A processor is used to execute the program, wherein the program is specifically used to: Acquire first information, where the first information consists of nodes with an associated relationship; Converting the first information into target information according to a conversion condition of the information format; wherein the conversion condition matches a data processing format of a pre-created evaluation model; Processing the target information using a pre-created evaluation model to obtain evaluation information corresponding to the first information, wherein the evaluation model is a neural network model obtained by training sample information, the sample information is obtained by converting log data, the nodes of the log data have a corresponding relationship with the nodes of the first information, and the log data at least includes the evaluation information corresponding to the first information; outputting the evaluation information; The converting of the first information into target information according to the information format conversion condition includes: Matching a node label corresponding to the information format for each node of the first information; Generate target information according to the node label.
Citation Information
Patent Citations
Method and device for cross-type conversation, apparatus, and computer-readable storage medium
CN109033223A
A processing flow adjusting method and device
CN109726002A
