CN111177389A - NLP technology-based classification method, system and storage medium for power charge notification and customer appeal collection - Google Patents
NLP technology-based classification method, system and storage medium for power charge notification and customer appeal collection Download PDFInfo
- Publication number
- CN111177389A CN111177389A CN201911395991.7A CN201911395991A CN111177389A CN 111177389 A CN111177389 A CN 111177389A CN 201911395991 A CN201911395991 A CN 201911395991A CN 111177389 A CN111177389 A CN 111177389A
- Authority
- CN
- China
- Prior art keywords
- classification
- customer service
- electric charge
- work order
- service work
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
- G06F16/353—Clustering; Classification into predefined classes
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/01—Customer relationship services
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Economics (AREA)
- Databases & Information Systems (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Tourism & Hospitality (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- General Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Primary Health Care (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Animal Behavior & Ethology (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- Finance (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention belongs to the electric power data processing technology, and relates to a classification method, a system and a storage medium for informing and urging customer appeal based on an NLP (non line segment process) technology, wherein the method comprises the following steps: sorting a classification knowledge map, building a classification frame meeting actual requirements, manually classifying and labeling the electricity charge customer service worksheets, and extracting classification rules through iterative operation; text cleaning is carried out on the electricity fee customer service work order to be classified; constructing a professional word bank to perform text word segmentation on the cleaned electric charge customer service work order; expressing a text vector of the electric charge customer service work order by using a TF-IDF algorithm; screening effective characteristics of the text vector by adopting an information gain method; and classifying the work orders by adopting an SVM (support vector machine) algorithm. The method can accurately divide words of the complaint contents, calculate the optimal classification of the text, realize the automatic classification of the complaint contents of the customers in the customer service electric charge type customer service work order, and solve the problems of low granularity of manual classification, excessive consumption of human resources, lagging classification speed and the like.
Description
Technical Field
The invention belongs to the technical field of power data processing, relates to machine learning, NLP and client work order classification, and particularly relates to a classification method, a classification system and a storage medium for power charge notification and client appeal urging based on an NLP technology.
Background
In order to seriously implement marketing work deployment and deeply promote marketing big data application work, a classification model of a client for an electric charge notice and an urging payment is constructed by taking the client as a center and based on 95598 electric charge work order basic data and utilizing big data processing technologies such as natural language processing, machine learning and the like, the appeal of the client for the electric charge notice and the feedback of the electric charge urging payment are focused in real time, the content of the electric charge notice and grouping differentiation are optimized, an electric charge urging payment characteristic group is identified, a proper urging payment strategy is formulated, the electric charge urging payment efficiency and effect are improved, and 95598 telephone traffic and complaint amount are reduced.
With the development of market economy and the continuous deepening of power system innovation, the service requirements of customers and social circles on power enterprises are higher and higher, and the power enterprises need to establish brand-new enterprise images by using a brand-new service face and a management concept aiming at high-quality service, so that the market is won and the development is promoted. Improving the service level requires further improvement of the customer experience and customer satisfaction. In order to improve the customer satisfaction, the customer dissatisfaction point is required to be started, and customer service work order appeal analysis is a key link.
The traditional complaint handling and analyzing method is to manually classify the complaints according to the complaint classification during complaint acceptance and analyze the complaint content texts one by one, and the method has the following defects:
(1) the classification of complaint acceptance is inaccurate: the acceptance staff accepts the complaints, and when classifying the complaints, the complaints are understood to be different in the content of the complaints of the clients, so that the classification is inaccurate, and the complaint processing judgment difficulty and the statistical data deviation are increased.
(2) Complaint hotspot capture is difficult: the hot spots are not found timely, are not clear and are difficult to focus.
(3) Content analysis is difficult: the method has the advantages of large data volume, more characters, complex processing process, high labor cost and incomplete hotspot analysis, and is difficult to find out the root cause of complaints, and the text content needs to be analyzed manually one by one.
(4) The analysis efficiency is low: due to the fact that the work order text amount is large, work order analysis time is long.
(5) The work is difficult to generate force: the timeliness and accuracy of complaint hotspot and difficult point identification are not high, penalty assessment is difficult to implement, key complaints are difficult to control, analysis results are difficult to apply, and the problem cannot be fundamentally solved.
In view of the above difficulties, and due to the characteristics of unstructured text content of the complaint work order, it is difficult to classify through simple analysis, so that the method for researching text mining intelligent classification becomes a key problem to be solved at present.
Disclosure of Invention
In order to solve the technical problems in the prior art, the invention provides a classification method, a system and a storage medium for customer appeal based on NLP technology, which are used for notifying and urging collection of electric charges, accurately segmenting the content of an electric charge customer service work order, vectorizing the segmented words, performing classification modeling on the electric charge customer service work order, calculating the optimal classification of texts, realizing automatic classification of the customer complaint content in the electric charge customer service work order, and solving the problems of low classification granularity, excessive consumption of human resources, lagged classification speed and the like in the current manual classification.
The invention relates to a classification method of power charge notification and customer appeal urging based on NLP technology, which comprises the following steps:
s1, combing the classification knowledge map, and building a classification frame according with actual needs; classifying the electricity fee customer service order manually based on a classification frame, analyzing the reason of customer complaints, labeling a historical electricity fee customer service order, and extracting a classification rule through iterative operation of a classification model;
s2, text cleaning is carried out on the electric charge customer service work order to be classified;
s3, constructing a professional lexicon to perform text word segmentation on the cleaned electric charge customer service work order;
s4, representing the text vector of the electric charge customer service work order by using a TF-IDF algorithm;
s5, based on the information theory, screening the effective characteristics of the text vector by using an information gain method;
and S6, classifying the work orders by adopting an SVM (support vector machine) algorithm.
Preferably, in step S3, based on the power grid industry professional vocabulary, a professional vocabulary library is manually constructed, and a word segmentation algorithm is combined, so as to extract and supplement the professional vocabulary lacking in the corpus from the electric charge customer service work order by human experience, thereby realizing accurate word segmentation of the unknown words.
Preferably, the vector conversion of the Chinese words is performed by using the TF-IDF method in step S4, and the process is as follows:
the frequency of occurrence of a certain entry in a certain electric charge type customer service work order is more than the number of words of the electric charge type customer service work order, and the number is used as a word frequency TF normalization formula:
calculating the frequency IDF of the reverse document, and obtaining words which appear much in a certain type of electric charge type customer service work order and appear little in other electric charge type customer service work orders as the theme of the electric charge type customer service work order:
and finally, calculating the word frequency-reverse document frequency as follows:
TF-IDF＝TF*IDF
therefore, the TF-IDF value of each vocabulary feature in each electric charge type customer service work order is calculated, the assignment of the vocabulary feature is completed, and finally the word vectorization process of the electric charge type customer service work order is completed.
The invention relates to an electric charge notification and client appeal collection classification system based on an NLP technology, which comprises:
the system comprises a classification knowledge map and historical electric charge customer service work order marking module, a classification framework and a data processing module, wherein the classification knowledge map and the historical electric charge customer service work order marking module are used for carding the classification knowledge map and building the classification framework which meets the actual requirement; classifying the electricity fee customer service order manually based on a classification frame, analyzing the reason of customer complaints, labeling a historical electricity fee customer service order, and extracting a classification rule through iterative operation of a classification model;
the text cleaning module is used for performing text cleaning on the electricity fee customer service work orders to be classified;
the text word segmentation module is used for constructing a professional word bank to perform text word segmentation on the cleaned electric charge customer service work order;
the text vector conversion module is used for representing the text vector of the electric charge customer service work order by using a TF-IDF algorithm;
the feature screening module is used for screening effective features of the text vectors by adopting an information gain method based on an information theory;
and the classification module classifies the work orders by adopting an SVM (support vector machine) algorithm.
Preferably, the text vector conversion module performs Chinese word vector conversion by using a TF-IDF method, and the process is as follows:
the frequency of occurrence of a certain entry in a certain electric charge type customer service work order is more than the number of words of the electric charge type customer service work order, and the number is used as a word frequency TF normalization formula:
calculating the frequency IDF of the reverse document, and obtaining words which appear much in a certain type of electric charge type customer service work order and appear little in other electric charge type customer service work orders as the theme of the electric charge type customer service work order:
and finally, calculating the word frequency-reverse document frequency as follows:
TF-IDF＝TF*IDF
therefore, the TF-IDF value of each vocabulary feature in each electric charge type customer service work order is calculated, the assignment of the vocabulary feature is completed, and finally the word vectorization process of the electric charge type customer service work order is completed.
The storage medium according to the invention has stored thereon computer instructions which, when executed by a processor, carry out the steps of the method according to the invention.
Compared with the prior art, the invention has the following advantages and beneficial effects:
1. the method comprises the steps of combing historical electric charge customer service work order data, analyzing main complaint tendency of a client, combing out a complaint attribution frame, labeling the historical electric charge customer service work order based on the attribution frame, adopting a regular expression technology to clean a text, building a special word stock of participles, combining a Chinese natural language word segmentation technology to realize accurate word segmentation of complaint contents, vectorizing the segmented words through an improved TF-IDF algorithm, utilizing a support vector machine algorithm suitable for text classification to perform classification modeling on the electric charge customer service work order, calculating the optimal classification of the text, realizing automatic classification of the complaint contents in the electric charge customer service work order, and solving the problems of low precision of classification degree, excessive consumption of human resources, lagging classification speed and the like in the current manual classification.
2. The method and the system intelligently analyze the text content of the power bill demand, study the hot demand of the client for power bill notification and collection, group the clients with different types of demands based on the text classification result, and provide accurate guidance decision for improving the satisfaction degree of the client. By expanding the service area, passive service is converted into active service, customer satisfaction and the image of the power enterprise are improved, and enterprise competitiveness and continuous development capability are further improved.
Drawings
FIG. 1 is a schematic flow chart of the classification method of the present invention;
FIG. 2 is a diagram of constructing a specialized thesaurus for word segmentation;
fig. 3 is a process diagram of text cleaning, text word segmentation and word vectorization.
Detailed Description
The present invention will be described in further detail with reference to examples and drawings, but the present invention is not limited thereto.
Examples
The method comprises the steps of combing historical electric charge customer service order data, analyzing core appeal of a client and combing out a complaint attribution frame, labeling responsibility departments, professional classifications, complaint events and error points of the historical electric charge customer service order based on the attribution frame, cleaning texts by adopting a regular expression technology, removing words such as greetings and idioms which do not contribute to classification, realizing accurate word segmentation of complaint contents by building a word segmentation professional lexicon and combining a Chinese natural language word segmentation technology, creatively vectorizing the segmented words by an improved TF-IDF algorithm, carrying out classification modeling on the electric charge customer service order by a support vector machine algorithm suitable for text classification, analyzing the optimal classification of the texts, and realizing automatic classification of the complaint contents of the client in the electric charge customer service order. The invention relates to a classification method of power charge notification and customer appeal urging based on NLP technology, which comprises the following steps:
s1, combing the classification knowledge map, and labeling the historical electric charge customer service work order;
(1) classification-knowledge-graph overview
The classification knowledge map is a classification framework formed by connecting all levels of classification levels of the electric charge customer service work order. In this embodiment, each electricity fee customer service work order corresponds to four classes, which are: department of responsibility, professional classification, complaint events, error points. The error points can be divided into service error points and service error points, and the service error points are identified by a suffix _Kwhen the classification knowledge graph is carded. By carrying out four-level classification on the electric charge customer service worksheet, the responsibility department to which the complaint belongs, the professional classification of the service corresponding to the department, the specific reasons and the optimization direction of the service workflow and the appeal event can be quickly known.
(2) Categorised knowledge map is combed
The historical classification frame is automatically added by different salesmen according to self cognition when classifying work orders, so that the classification is disordered, the granularity of classification of the same class is inconsistent, and the logic between different classes is not strict. The reasonable construction of the classification frame is the key of text classification application, and whether the classification result is helpful for understanding the current situation of complaints and is applied to the improvement of service quality is determined. The embodiment is based on abundant business experience of business experts, and a classification frame which meets actual needs is built by combining business requirements; carding as follows:
and (4) responsibility departments: marketing and operation inspection;
professional classification: voltage quality, emergency maintenance service, power supply quality, reading, checking, charging, power utilization change, business hall, business expansion, metering, power construction, intelligent charging, power failure problem, rural power grid transformation, customer information, power supply facility, intelligent power, frequent power failure, inspection, wind running, distribution network transformation, national power grid customer service, others, electronic invoice, quick response, motorcade and the like;
appeal event: emergency repair, frequent power failure, charge urging, frequent power failure, time-of-use electricity price, billing, new installation, meter/meter box replacement, meter counting, pole moving and line changing, power construction, temporary power utilization, voltage quality, emergency repair, construction site non-recovery, intelligent charge opening, intelligent charge canceling, no-fault power failure, consultation power failure, charging mode, construction damage compensation, meter rotation, house change, meter inspection, line rectification, blue and green compensation, power failure and power restoration according to plan, business hall service, short message, circuit removal, division, capacity increase, meter reading, charging pile, meter/meter box removal, telephone consultation, line/equipment hidden danger processing, meter installation, electricity connection, electricity price execution, customer information change, civil compensation, class change, payment, expense sale, power failure information announcement, matching power failure and power failure, household appliance damage checking, electricity stealing, photovoltaic compensation, inquiry information, power consumption information, and the like, Claim for a customer, out of duty, meter change, capacity reduction, electricity fee refund, waste disposal, receiving private activities, quality of first-aid repair, low voltage handling, power facility maintenance, claims, refund, customer information change, rural power grid modification, meter box repair, marketing activities, household meter modification, personal injury compensation processing, business handling, meter closing, new installation survey, troubleshooting, installation location migration, electricity fee consultation, supply and utilization contract, business, meter box installation, site survey, account sale, photovoltaic power generation, electricity fee distribution, credit issue, safe distance, outage of distribution room air conditioning, site consultation, warning board setting, meter box modification, power outage without permission, customer property equipment theft, credit exchange, noise processing, private power connection, others, default funds, line change, line release, electricity fee standard, power restoration, business handling, transformer replacement, printing list, electricity consumption standard, power return, electricity return, power consumption refund, electricity consumption balance, and electricity consumption balance, Abnormal charge, abnormal electric quantity check, power supply capacity, damaged and temporarily disassembled client property right meter boxes, untimely wiring after matched disassembly, early warning threshold value change, account number binding, line erection, palm power downloading, electricity stealing processing, non-working time client contact, ammeter resetting, suspension, field instruction meter reading, door service, matched power failure, vehicle dispute, report verification, special transformer power failure non-notification, telephone service and the like;
error point: voltage low, service attitude difference _ K, frequent power failure, electric charge rejection, refusal/deniable acceptance _ K, refusal of issuing an electric charge invoice, material non-specification, indiscriminate charging, meter line connection error, answer client non-cashing _ K, client loss non-processing _ K, refusal of acceptance, change intelligent payment non-notification, frequent power failure, non-processing, voltage low, service attitude difference _ K, no-fault power failure, client loss non-processing _ K, answer client non-cashing _ K, list change non-notification, change account name non-notification, pending determination, meter line connection error, green claim non-processing, delayed power transmission, designated construction party _ K, arrival over time limit, non-fault power failure notification, abnormal operation _ K, indiscriminate charging, repeated error short message sending, electric charge rejection, field error charging, non-invoice-providing _ K, client loss non-processing, power failure in advance, material non-specification, No arrearage is interrupted, acceptance is refused, zero electricity is used, overtime is handled, meter _ K is removed without authorization, charging notice is not standard _ K, no notice is given without authorization to pull wire, time-of-use price is opened and not executed, client article is not returned _ K, waste is not cleaned, paper error-hastening fee, client information change is not informed, power failure information notice is not accurate, charging is disorderly _ K, power restoration overtime, capacity reduction and failure-free power failure are not reduced, normal operation is not performed, power failure information notice is not provided, account name is not changed, overtime, photovoltaic power generation supplement is not in place, imperfect _ K is processed, error and error are copied, account is not in duty _ K, other _ K, power failure is delayed, nonstandard handling is provided, invoice is provided, private _ K is connected, account is not informed, electricity fee is mistakenly deducted, account name is mistakenly registered, meter is checked, acceptance is refused/pushed overtime _ K, client information leakage is high, voltage is not notified, meter is not removed, and not, Requesting property _ K, failure handling is not thorough, power transmission is advanced, new installation is over time, reply error _ K, short message error-prompting fee, no change of contact telephone, bad repair quality, unstable voltage, no refund deposit, charging error, refusing to make electric charge invoice, telephone error-prompting fee, failure notice _ K, return receipt and fact disagreement _ K, inaccurate power transmission time reply, sending short message according to requirement, material notice not normalized _ K, circuit dismantling not notice _ K, electric charge notice error, table dismantling not notice _ K, file information error/incompletion, service incompletion _ K, unchanged client contact telephone, unauthorized taking off client table offline _ K, forced wiring, unauthorized moving table _ K, wrong user number notice, material not notified once, failure handling, unauthorized change day, wiring error, nuclear damage overtime, loss overtime, failure handling date, failure handling, failure, Non-surveying, non-performing survey, manual hiring _ K, optional pull-off _ K, damaged client _ K, bribery _ K, improper charging mode _ K, inaccurate power failure notification, small replacement line meter offline notification, business non-acceptance, address registration error, power failure _ K, construction damage compensation, optional reply _ K, non-standard telephone service _ K, important card _ K, mandatory transaction _ K, private power connection _ K, non-transmission of power failure notification short message, unexecuted meter-closing electricity price, non-meter-installation, others, private-connected client property wire, unchanged client information, power connection failure, inconsistent reply _ K, non-in-place subsidy, drainage to client garage _ K, unprocessed _ K, designated team processing client transfer meter box business _ K, meter-changing leakage, non-meter-installation, non-standard transaction _ K, cost control early warning value modification failure, charging control early warning value modification, Binding error of payment treasures, line construction without permission, no change in charge, preferential activity enjoying, paying electric charge, repeated charge, compensation delivery error, non-working time connection client _ K, designated construction unit _ K, failure processing _ K, meter reading date change, failure to bill, account number notification error, information notification error, power failure information release channel, poor installation quality, no-person service _ K, failure of timely power failure information report, no-specified door service _ K, power charge calculation error, agricultural quality difference, failure of intelligent payment audit, civil compensation, too high early warning amount, failure of line change trend notification _ K, failure of power failure short message notification, failure of printing, loss of client material _ K, failure of electronic invoice reimbursement, failure of business hall order specification _ K, failure of meter removal _ K, limb conflict _ K, failure of time sharing electric price opening notification, failure of time sharing electric price notification, failure of printing, failure of payment due payment, The electricity fee issuing is wrong, the payment cannot be made, the electricity fee is generated when the suspension and recovery is not applied, the power failure non-notice is cancelled, the client _ K is scared, the cash is rejected, the refund is not paid, the installation fee is not refunded, and the like.
(3) Sample classification labeling
The text classification model adopts a core technology of machine learning algorithm, a model building stage is required to manually classify the electric charge customer service work order based on a classification frame, an accurate work order label is marked on the electric charge customer service work order, the machine learning algorithm analyzes the work order which is classified historically, and classification rules are extracted through iterative operation of the classification model. And in the classification model application stage, applying the learned rules to other unclassified work orders to realize automatic classification.
S2, text cleaning is carried out on the electric charge customer service work order to be classified by using a regular expression;
the work of text cleaning is indispensable, can effectively reduce the vocabulary noise, remains more effective text characteristic, obtains better text characteristic for classification model reaches higher precision. The specific method comprises the following steps:
(1) removing punctuation marks
Punctuation is removed because it does not add any additional information in the text data. Therefore, deleting all symbols can help to reduce the size of training data and improve the training performance of the model.
(2) Removing stop words and rare words
Stop words are words whose information does not help the classification of the model or even leads to a certain misleading and which should be deleted from the text data. In the embodiment, prepositions, greetings and other vocabularies in the electric charge customer service work order are collected to create the disabled word bank, and corresponding vocabularies in the complaint text are removed according to the disabled word bank, so that the purpose of cleaning the text is achieved.
The rare words refer to words that exist only in a few work orders. Due to the rarity of the low-frequency vocabulary, the improvement on the model performance is extremely limited, and the rare words can be replaced by other synonyms to improve the word frequency, or the words can be directly deleted to improve the iteration efficiency of the model.
(3) Disambiguation translation
Disambiguating some homophonic wrongly written words in the textual description, for example: the words such as 'accompanying product' and 'indemnity' are disambiguated and converted into 'indemnity'.
(4) Personalized text conversion
Texts like house numbers, work order numbers, telephone numbers and the like have no great difference among different numbers, and the model only focuses on the meaning represented by the numbers and does not focus on the specific numerical values of the numbers. In the embodiment, through analyzing the electric charge type customer service work order data of the 95598 hotline, the specific numerical values of the house number, the work order number and the telephone number are respectively replaced by the 'house number, the work order number and the telephone number', text features with thousands of inconsistent costs are uniformly expressed as the same text features, and the performance of the model can be improved.
(5) Removing idioms
Similar to customer incoming call reflection and power company timely processing, the texts are general texts of different text categories, the word frequency is high, but the classification is not assisted, and the words should be removed so as to improve the model iteration efficiency.
S3, constructing a professional lexicon to perform text word segmentation on the cleaned electric charge customer service work order, and improving the accuracy of text word segmentation;
text segmentation is a special and important link in the Chinese text classification technology. Since words are the smallest language units that can be used independently, and chinese text is not like english, there is no display mark such as any space between chinese words and words to indicate the boundaries of words. Therefore, text word segmentation is a basic link of the Chinese natural language processing technology, the text word segmentation effect is good or bad, and the performance of a subsequent text classification model is determined.
Chinese word segmentation needs to utilize Chinese automatic word segmentation packages, such as jieba, HanLP and the like, and all the word segmentation packages adopt a statistical method based on large-scale training corpora. Words and phrases which are contained in the electric charge customer service work order but not contained in the corpus are called unknown words, and the unknown words can not be correctly separated by the word segmentation packet. The power grid is a highly professional industry, and contains a large number of unknown words which do not exist in an external corpus, such as 'apportionment of electric quantity', and the like.
The invention is based on the professional vocabulary of the power grid industry, extracts and supplements the professional vocabulary lacking in the corpus from the electric charge customer service work order by manually constructing the professional vocabulary bank and combining the word segmentation algorithm and by means of human experience so as to realize the accurate word segmentation of the unknown words and solve the problem that the professional vocabulary amount of the traditional corpus is insufficient. These artificially constructed dictionaries are continuously collected and stored. In this embodiment, 10000 customer service work orders are analyzed, approximately 1000 unregistered words such as young plants, photovoltaic power generation, personnel's peculiarities, payment urging, business expansion change and the like are collected from the analyzed customer service work orders, and a constructed professional lexicon is shown in fig. 2, so that the word segmentation accuracy and the classification effect of the models are improved.
S4, representing the text vector of the electric charge customer service work order by using a TF-IDF algorithm;
a text is represented as a string of characters and punctuation marks, but a computer cannot recognize the string and needs to be converted into numbers for processing. In order to make a computer efficiently process real text, an ideal formal representation must be found, which reflects the content of a document in reality and has the ability to distinguish different documents. The embodiment adopts a word vector method to represent the text by a vector. When a word vector method is adopted for text representation, two steps of establishing a feature dictionary and assigning features are required: generating a characteristic item sequence required by text representation according to the training sample set; and according to the characteristic item sequence of the text, carrying out weight assignment, normalization and other processing on each electric charge customer service work order in the training text set and the test sample set, and converting the electric charge customer service work orders into characteristic vectors required by a machine learning algorithm.
The feature weight is used for measuring the importance degree or the distinguishing capability of a certain feature item in the document representation. The general method of weight calculation is to give a certain weight to the feature items by using the statistical information of the text, mainly the word frequency. The TF-IDF method measures the weight of a certain feature from two indexes of word frequency and inverse document frequency. The TF-IDF (term frequency-inverse document frequency) algorithm is a statistical method for evaluating the importance of a word to a power charge type customer service work order, wherein the importance of the word is increased in proportion to the occurrence frequency of the word in a file, but is decreased in inverse proportion to the occurrence frequency of the word in all the work orders. The more times a word appears in a document and the less times it appears in all other documents, the more representative the content of the document the word will be and the greater the classification will be. For example, the word like "hello" appears in many work orders, is not unique to a certain class of workers, has little importance, and has little corresponding TF-IDF value; the terms of the business hall and the meter box are more in the responsibility department-marketing class, and the frequency of the responsibility department-operation class is less, the two terms of the business hall and the meter box can represent the content of the work order, and the TF-IDF of the two terms is larger. Different from the traditional TF-IDF mode, the invention improves the TF-IDF algorithm on the basis of the original method, so that the classification effect of the model is better. The invention uses TF-IDF method to convert Chinese word vector, which comprises the following steps:
TF represents word frequency, which refers to the frequency of occurrence of a certain specified word in a document, and the frequency of occurrence of the same word in a long work order is more than that of occurrence of the same word in a short document; in order to avoid the word frequency biased to the long work order, in this embodiment, the number of times that a word (i.e., a certain entry in a certain work order) appears is required to be used as a word frequency TF normalization formula to prevent the word frequency biased to the long work order, and the calculation formula is as follows:
IDF denotes the inverse document frequency. Some common words appear in each work order in a large amount, for example, the word "is calculated by using a TF formula, the weight calculated by using the TF formula is large, but the word cannot reflect the theme of one work order, in order to effectively distinguish different work orders, the words which appear much in one kind of work orders and few in other work orders are needed, the words can reflect the theme of the work order, the purpose cannot be realized by the word frequency TF, and the reverse document frequency IDF can realize the purpose. If the work order containing a certain word is less, the reverse document frequency IDF is larger, and the word has good category distinguishing capability. The calculation formula of the inverse document frequency IDF is as follows:
high frequency words in a certain work order, and low file frequency of the words in the whole document set can generate TF-IDF values with high weight, namely TF-IDF tends to filter out common words and keep important words. And finally, the calculation formula of the TF-IDF is as follows:
TF-IDF＝TF*IDF
in the embodiment, the TF-IDF value of each vocabulary feature in each electric charge customer service work order is calculated, so that the assignment of the vocabulary features is completed, and finally the word vectorization process of the text work order is completed.
The process of text cleaning, text word segmentation and word vectorization is shown in fig. 3.
S5, based on the information theory, screening effective characteristics by using an information gain method;
after word vectorization, more than 1 ten thousand of features used for representing text vectors are available, and the number of features is too large, which may cause dimension disasters and influence the model efficiency; therefore, effective features need to be screened out to be reserved, and irrelevant features need to be removed to improve the model effect.
The traditional characteristic selection mode is that upper and lower limits are set based on word frequency, and words with too high or too low word frequency are removed. This is simple and easy to implement and does remove a portion of the noise. However, this method is only a borrowing algorithm, and the theory is not enough. According to the information theory, some features, although having low frequency of appearance, often contain more information and are of great importance for classification. For such features, the term frequency method should not be used to exclude them directly from the vector features.
The invention starts from the information theory and adopts an information gain method to screen the effective characteristics of the text vector. The information gain method measures the importance degree of a characteristic item according to the amount of information which can be provided by the characteristic item for the whole classification, thereby determining the choice of the characteristic item. The information gain of a certain feature item refers to the difference of the information quantity provided for the whole classification when the feature exists or does not exist, the information quantity is measured by the entropy, the difference between the entropy of the work order without considering any feature and the entropy of the work order with considering the feature is as follows:
wherein, P (C)j) Is represented by CjProbability of occurrence of class sheet in all work sheets, P (t)i) Indicating that all work orders contain the characteristic item tiProbability of work order of (1), P (C)j|ti) Representing the inclusion of characteristic items t in the work orderiWhen it belongs to CjThe conditional probability of a class is determined,
indicating that the work order does not contain the characteristic item tiThe probability of the work order of (a),
indicating that the work order does not contain the characteristic item tiWhen it belongs to CjThe conditional probability of a class, M, represents the number of classes.
A characteristic information gain describes the amount of information it contains that can help predict the class attributes. In practical application, because the frequency of occurrence of many features with higher information gain is often lower, when the number of features selected by using the information gain is less, the problem of data sparseness exists, and the classification effect is also poor. Therefore, the invention carries out improvement processing when screening the characteristics, calculates the information gain of each word appearing in the electric charge customer service work order, sets a threshold value, removes the entries with the information gain lower than the set threshold value from the characteristic space, and then selects the characteristics according to the sequence of the gain values from high to low to form the characteristic vector. By this method, over 1000 features are ultimately retained, reducing the invalid features by 90%.
And S6, classifying the work orders by adopting an SVM (support vector machine) algorithm, and improving the performance of the model.
The extraction of the classification rules finally depends on the operation of the algorithm, and the selection of the algorithm is particularly important. In the traditional method, a naive Bayes classifier is adopted more, but the prior probability is used as an important parameter in the algorithm, and customer service complaints are very seasonal, and if frequent power failure occurs, the complaints in summer are high in proportion, the complaints in winter are reduced, the prior probability changes along with time, and the naive Bayes algorithm is not suitable.
The invention adopts the SVM support vector machine algorithm with excellent classification performance. The basic idea of SVM is to find a decision plane in vector space that best partitions the data points in the two classes. The support vector machine classification method is to find a decision plane with the largest classification interval in a training set. Due to the characteristics, the SVM has good classification performance and model generalization capability. The support vector machine can be formalized into a problem for solving convex quadratic programming through a learning strategy of interval maximization, which is equivalent to a minimization problem of a regularized hinge loss function. The support vector machines include linear branched support vector machines, linear support vector machines, and non-linear support vector machines. When the training data is linearly divisible, learning a linear classifier through hard interval maximization, namely a linear branch support vector machine, which also becomes a hard interval support vector machine; when the training data is approximately linearly divisible, a linear classifier, namely a linear support vector machine, also called a soft-interval support vector machine, is also learned through soft-interval maximization; when the training data is linearly independent, the nonlinear support vector machine is learned through the kernel skill and soft interval maximization. Theoretically, the effect of the nonlinear support vector machine using kernel technique is not lower than that of the linear support vector machine, but the nonlinear support vector machine needs to go through a complicated process of adjusting kernel function parameters. In practical application, the embodiment uses a linear support vector machine.
Specifically, the input space and the feature space are assumed to be two different spaces. The input space is an Euclidean space or a discrete set, and the feature space is an Euclidean space or a Hilbert space. The linear branching support vector machine assumes a one-to-one correspondence of the elements of these two spaces and maps the input in the input space to a feature vector in the feature space. The nonlinear support vector machine maps the input into feature vectors using a nonlinear mapping from the input space to the feature space. The learning goal of the support vector machine is to find a separating hyperplane in the feature space, which can divide the work order into different classes. The separating hyperplane corresponds to the equation wx + b ═ 0, which is determined by the hyperplane coefficient w and the intercept b. The separating hyperplane divides the feature space into two parts, one part being a positive class and one part being a negative class. One side of the normal vector points to is a positive class, and the other side is a negative class. In general, when training data sets are linearly separable, there are infinite separation hyperplanes that can correctly separate the two types of data. The perceptron utilizes the strategy of minimum misclassification to obtain a separation hyperplane, but the solutions at this time are infinite. The linear branching support vector machine uses interval maximization to solve the optimal separation hyperplane, and the solution is unique.
The basic idea of support vector machine learning is to solve separate hyperplanes that can correctly partition the training data set and are maximally geometrically spaced. For a linearly separable training data set, there are an infinite number of linearly separable separation hyperplanes, but the most geometrically spaced separation hyperplane is unique. The intuitive explanation for interval maximization is: finding the hyperplane with the largest geometric separation on the training data set means that the training data is classified with a sufficiently large degree of certainty. That is, not only are the positive and negative instance points separated, but there is sufficient confidence that the most difficult instance points are separated. Such hyperplanes should have good classification prediction capabilities for new instances that are unknown. In order to maximize the interval, it is necessary to find the data points with the minimum interval, which are the support vectors. Finding the data point of the minimum interval, it is necessary to maximize the interval, which is formulated as:
where y represents the true classification of the work order and w, b are the coefficients and intercept of the hyperplane.
Solving this formula requires fixing one of the factors and maximizing the other factor. That is, let the function interval of all support vectors be label × (w)Tx + b) are all 1, the final solution can be obtained by solving the maximum of the reciprocal of | | w |. However, not all data points have a functional separation equal to 1, only those closest to the hyperplane have a value of 1, while data points further from the hyperplane must have a functional separation greater than 1, which is about the support vector machineThe beam condition. For such optimization problems, lagrange multiplier method can be used. By introducing lagrange multipliers, the original problem can be expressed based on constraints. Because of the inevitable noise in the data, slack variables are introduced into the support vector machine to allow some data points to be on the wrong side of the separation plane, so that the optimization goal is not changed, and only the constraint condition is influenced. The relaxation variable is used to control the weight of these two objectives of "maximize interval" and "guarantee that most points are functionally spaced less than 1.0". In optimizing the classification model, the slack variable is a hyper-parameter and needs to be adjusted in constant attempt. After final lagrangian transformation, the specific optimization objective can be written as:
the constraint conditions are as follows:
where y represents the true classification of the work order, α { α }1,…αiThe constant C is the regularization coefficient.
determining the above-mentioned alphaiAnd obtaining the SVM support vector machine model.
Inputting the text data converted into word vectors into an SVM algorithm, adopting K-fold cross validation, continuously iterating the model, and adjusting various parameters of the model, such as loss functions, penalty term coefficients and the like, according to the classification accuracy of the model. And finally determining the optimal parameters of the model.
When the technical scheme adopted by the invention is used for solving the problems in the prior art, the technical scheme can also be a classification system for informing the electricity charge and urging the customer to appeal based on an NLP technology, and the classification system comprises:
the classification knowledge map and historical electric charge customer service work order marking module is used for realizing the step S1, combing the classification knowledge map and building a classification frame meeting the actual requirement; classifying the electricity fee customer service order manually based on a classification frame, analyzing the reason of customer complaints, labeling a historical electricity fee customer service order, and extracting a classification rule through iterative operation of a classification model;
the text cleaning module is used for realizing the step S2 and performing text cleaning on the electricity fee customer service work order to be classified;
the text word segmentation module is used for realizing the step S3, and constructing a professional word bank to perform text word segmentation on the cleaned electric charge customer service work order;
the text vector conversion module is used for realizing the step S4, and expressing the text vector of the electric charge customer service work order by using a TF-IDF algorithm;
the characteristic screening module is used for realizing the step S5, and screening the effective characteristics of the text vector by adopting an information gain method based on an information theory;
and the classification module is used for realizing the step S6 and classifying the work orders by adopting an SVM (support vector machine) algorithm.
The technical solution of the present invention can also be embodied as a storage medium on which computer instructions are stored, and the computer instructions, when executed by a processor, implement the steps of the classification method for electric charge notification and customer appeal solicitation of the present invention.
The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.
Claims (10)
1. The classification method for notifying the electricity charge and receiving the customer appeal based on the NLP technology is characterized by comprising the following steps of:
s1, combing the classification knowledge map, and building a classification frame according with actual needs; classifying the electricity fee customer service order manually based on a classification frame, analyzing the reason of customer complaints, labeling a historical electricity fee customer service order, and extracting a classification rule through iterative operation of a classification model;
s2, text cleaning is carried out on the electric charge customer service work order to be classified;
s3, constructing a professional lexicon to perform text word segmentation on the cleaned electric charge customer service work order;
s4, representing the text vector of the electric charge customer service work order by using a TF-IDF algorithm;
s5, based on the information theory, screening the effective characteristics of the text vector by using an information gain method;
and S6, classifying the work orders by adopting an SVM (support vector machine) algorithm.
2. The classification method according to claim 1, wherein the classification knowledge map in step S1 is a classification framework formed by connecting classification levels of the electricity fee customer service work order; each electric charge customer service work order corresponds to four classes, which are respectively: department of responsibility, professional classification, complaint events, error points.
3. The classification method according to claim 1, wherein in step S3, based on the power grid industry professional vocabulary, the professional vocabulary library is constructed manually, and in combination with the word segmentation algorithm, the professional vocabulary lacking in the corpus is extracted and supplemented from the electricity fee customer service work order by human experience, so as to realize accurate word segmentation of the unknown words.
4. The classification method according to claim 1, wherein the TF-IDF method is used for the chinese word vector transformation in step S4, and the procedure is as follows:
the frequency of occurrence of a certain entry in a certain electric charge type customer service work order is more than the number of words of the electric charge type customer service work order, and the number is used as a word frequency TF normalization formula:
calculating the frequency IDF of the reverse document, and obtaining words which appear much in a certain type of electric charge type customer service work order and appear little in other electric charge type customer service work orders as the theme of the electric charge type customer service work order:
and finally, calculating the word frequency-reverse document frequency as follows:
TF-IDF＝TF*IDF
therefore, the TF-IDF value of each vocabulary feature in each electric charge type customer service work order is calculated, the assignment of the vocabulary feature is completed, and finally the word vectorization process of the electric charge type customer service work order is completed.
5. The classification method according to claim 1, wherein the step S5 is to calculate an information gain for each word appearing in the electricity fee type customer service order when filtering the valid features, set a threshold, remove entries whose information gain is lower than the set threshold from the feature space, and then select feature component feature vectors in order of gain value from high to low.
6. The classification method according to claim 1, wherein a linear support vector machine is used in step S6.
7. An electricity charge notification and customer appeal collection classification system based on NLP technology is characterized by comprising:
the system comprises a classification knowledge map and historical electric charge customer service work order marking module, a classification framework and a data processing module, wherein the classification knowledge map and the historical electric charge customer service work order marking module are used for carding the classification knowledge map and building the classification framework which meets the actual requirement; classifying the electricity fee customer service order manually based on a classification frame, analyzing the reason of customer complaints, labeling a historical electricity fee customer service order, and extracting a classification rule through iterative operation of a classification model;
the text cleaning module is used for performing text cleaning on the electricity fee customer service work orders to be classified;
the text word segmentation module is used for constructing a professional word bank to perform text word segmentation on the cleaned electric charge customer service work order;
the text vector conversion module is used for representing the text vector of the electric charge customer service work order by using a TF-IDF algorithm;
the feature screening module is used for screening effective features of the text vectors by adopting an information gain method based on an information theory;
and the classification module classifies the work orders by adopting an SVM (support vector machine) algorithm.
8. The classification system according to claim 7, wherein the classification knowledge map is a classification framework connected by classification levels of an electric charge customer service work order; each electric charge customer service work order corresponds to four classes, which are respectively: department of responsibility, professional classification, complaint events, error points.
9. The classification system according to claim 7, wherein the text vector conversion module performs the chinese word vector conversion by using a TF-IDF method as follows:
the frequency of occurrence of a certain entry in a certain electric charge type customer service work order is more than the number of words of the electric charge type customer service work order, and the number is used as a word frequency TF normalization formula:
calculating the frequency IDF of the reverse document, and obtaining words which appear much in a certain type of electric charge type customer service work order and appear little in other electric charge type customer service work orders as the theme of the electric charge type customer service work order:
and finally, calculating the word frequency-reverse document frequency as follows:
TF-IDF＝TF*IDF
therefore, the TF-IDF value of each vocabulary feature in each electric charge type customer service work order is calculated, the assignment of the vocabulary feature is completed, and finally the word vectorization process of the electric charge type customer service work order is completed.
10. Storage medium having stored thereon computer instructions, characterized in that the computer instructions, when executed by a processor, carry out the steps of the method according to any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911395991.7A CN111177389A (en) | 2019-12-30 | 2019-12-30 | NLP technology-based classification method, system and storage medium for power charge notification and customer appeal collection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911395991.7A CN111177389A (en) | 2019-12-30 | 2019-12-30 | NLP technology-based classification method, system and storage medium for power charge notification and customer appeal collection |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111177389A true CN111177389A (en) | 2020-05-19 |
Family
ID=70654355
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911395991.7A Pending CN111177389A (en) | 2019-12-30 | 2019-12-30 | NLP technology-based classification method, system and storage medium for power charge notification and customer appeal collection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111177389A (en) |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111737421A (en) * | 2020-08-07 | 2020-10-02 | 杭州六棱镜知识产权科技有限公司 | Intellectual property big data information retrieval system and storage medium |
| CN111753840A (en) * | 2020-06-18 | 2020-10-09 | 北京同城必应科技有限公司 | A business card ordering technology for intra-city logistics distribution |
| CN111783438A (en) * | 2020-05-22 | 2020-10-16 | 贵州电网有限责任公司 | Hot word detection method for work order analysis |
| CN111856209A (en) * | 2020-07-23 | 2020-10-30 | 广东电网有限责任公司清远供电局 | Power transmission line fault classification method and device |
| CN112183068A (en) * | 2020-09-30 | 2021-01-05 | 深圳供电局有限公司 | A method and system for differentiated handling of customer complaints |
| CN112667777A (en) * | 2020-12-28 | 2021-04-16 | 广东电网有限责任公司中山供电局 | Classification method for client incoming call appeal |
| CN112711947A (en) * | 2021-01-09 | 2021-04-27 | 国网湖北省电力有限公司电力科学研究院 | Text vectorization-based handling reference method in fault power failure repair work |
| CN112732918A (en) * | 2021-01-13 | 2021-04-30 | 国网山东省电力公司日照供电公司 | Complaint work order intelligent classification system |
| CN112991049A (en) * | 2021-04-13 | 2021-06-18 | 重庆度小满优扬科技有限公司 | Loan information processing method and electronic device |
| CN113254644A (en) * | 2021-06-07 | 2021-08-13 | 成都数之联科技有限公司 | Model training method, non-complaint work order processing method, system, device and medium |
| CN113284007A (en) * | 2021-05-27 | 2021-08-20 | 国网电力科学研究院武汉能效测评有限公司 | Power utilization information processing system based on power insurance package and processing method thereof |
| CN113434680A (en) * | 2021-06-29 | 2021-09-24 | 平安科技（深圳）有限公司 | User intention analysis method and device based on seat data and electronic equipment |
| CN113779008A (en) * | 2021-09-07 | 2021-12-10 | 杭州天宽科技有限公司 | Intelligent storage system for operating data of power intranet |
| CN115099768A (en) * | 2022-05-23 | 2022-09-23 | 远光软件股份有限公司 | Expense reimbursement method, device and equipment |
| CN115204822A (en) * | 2022-07-04 | 2022-10-18 | 中国电信股份有限公司 | Work order processing method and device and electronic equipment |
| CN116187676A (en) * | 2022-12-27 | 2023-05-30 | 中国电信股份有限公司 | A method, device, electronic device and storage medium for dispatching government affairs bills |
| CN116842165A (en) * | 2023-07-20 | 2023-10-03 | 中国银行股份有限公司 | Reason analysis methods, devices, equipment and media for negative customer service reviews |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101290626A (en) * | 2008-06-12 | 2008-10-22 | 昆明理工大学 | Text Classification Feature Selection and Weight Calculation Method Based on Domain Knowledge |
| CN107169086A (en) * | 2017-05-12 | 2017-09-15 | 北京化工大学 | A kind of file classification method |
| CN107908716A (en) * | 2017-11-10 | 2018-04-13 | 国网山东省电力公司电力科学研究院 | 95598 work order text mining method and apparatus of word-based vector model |
| KR20180120488A (en) * | 2017-04-27 | 2018-11-06 | 한양대학교 산학협력단 | Classification and prediction method of customer complaints using text mining techniques |
| CN109492091A (en) * | 2018-09-28 | 2019-03-19 | 科大国创软件股份有限公司 | A kind of complaint work order intelligent method for classifying based on convolutional neural networks |
-
2019
- 2019-12-30 CN CN201911395991.7A patent/CN111177389A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101290626A (en) * | 2008-06-12 | 2008-10-22 | 昆明理工大学 | Text Classification Feature Selection and Weight Calculation Method Based on Domain Knowledge |
| KR20180120488A (en) * | 2017-04-27 | 2018-11-06 | 한양대학교 산학협력단 | Classification and prediction method of customer complaints using text mining techniques |
| CN107169086A (en) * | 2017-05-12 | 2017-09-15 | 北京化工大学 | A kind of file classification method |
| CN107908716A (en) * | 2017-11-10 | 2018-04-13 | 国网山东省电力公司电力科学研究院 | 95598 work order text mining method and apparatus of word-based vector model |
| CN109492091A (en) * | 2018-09-28 | 2019-03-19 | 科大国创软件股份有限公司 | A kind of complaint work order intelligent method for classifying based on convolutional neural networks |
Non-Patent Citations (1)
| Title |
|---|
| 甄志龙: "《文本分类中的特征选择方法研究》", 31 December 2016 * |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111783438A (en) * | 2020-05-22 | 2020-10-16 | 贵州电网有限责任公司 | Hot word detection method for work order analysis |
| CN111753840A (en) * | 2020-06-18 | 2020-10-09 | 北京同城必应科技有限公司 | A business card ordering technology for intra-city logistics distribution |
| CN111856209A (en) * | 2020-07-23 | 2020-10-30 | 广东电网有限责任公司清远供电局 | Power transmission line fault classification method and device |
| CN111737421A (en) * | 2020-08-07 | 2020-10-02 | 杭州六棱镜知识产权科技有限公司 | Intellectual property big data information retrieval system and storage medium |
| CN112183068A (en) * | 2020-09-30 | 2021-01-05 | 深圳供电局有限公司 | A method and system for differentiated handling of customer complaints |
| CN112667777A (en) * | 2020-12-28 | 2021-04-16 | 广东电网有限责任公司中山供电局 | Classification method for client incoming call appeal |
| CN112711947A (en) * | 2021-01-09 | 2021-04-27 | 国网湖北省电力有限公司电力科学研究院 | Text vectorization-based handling reference method in fault power failure repair work |
| CN112711947B (en) * | 2021-01-09 | 2023-08-22 | 国网湖北省电力有限公司电力科学研究院 | A text vectorization-based method for dealing with power outages and emergency repairs |
| CN112732918A (en) * | 2021-01-13 | 2021-04-30 | 国网山东省电力公司日照供电公司 | Complaint work order intelligent classification system |
| CN112991049B (en) * | 2021-04-13 | 2023-05-30 | 重庆度小满优扬科技有限公司 | Loan information processing method and electronic equipment |
| CN112991049A (en) * | 2021-04-13 | 2021-06-18 | 重庆度小满优扬科技有限公司 | Loan information processing method and electronic device |
| CN113284007A (en) * | 2021-05-27 | 2021-08-20 | 国网电力科学研究院武汉能效测评有限公司 | Power utilization information processing system based on power insurance package and processing method thereof |
| CN113284007B (en) * | 2021-05-27 | 2023-07-04 | 国网电力科学研究院武汉能效测评有限公司 | Power consumption information processing system based on electric insurance package and processing method thereof |
| CN113254644B (en) * | 2021-06-07 | 2021-09-17 | 成都数之联科技有限公司 | Model training method, non-complaint work order processing method, system, device and medium |
| CN113254644A (en) * | 2021-06-07 | 2021-08-13 | 成都数之联科技有限公司 | Model training method, non-complaint work order processing method, system, device and medium |
| CN113434680A (en) * | 2021-06-29 | 2021-09-24 | 平安科技（深圳）有限公司 | User intention analysis method and device based on seat data and electronic equipment |
| CN113779008A (en) * | 2021-09-07 | 2021-12-10 | 杭州天宽科技有限公司 | Intelligent storage system for operating data of power intranet |
| CN113779008B (en) * | 2021-09-07 | 2024-05-24 | 杭州天宽科技有限公司 | Intelligent storage system for operation data of electric power intranet |
| CN115099768A (en) * | 2022-05-23 | 2022-09-23 | 远光软件股份有限公司 | Expense reimbursement method, device and equipment |
| CN115204822A (en) * | 2022-07-04 | 2022-10-18 | 中国电信股份有限公司 | Work order processing method and device and electronic equipment |
| CN115204822B (en) * | 2022-07-04 | 2025-10-17 | 中国电信股份有限公司 | Work order processing method and device and electronic equipment |
| CN116187676A (en) * | 2022-12-27 | 2023-05-30 | 中国电信股份有限公司 | A method, device, electronic device and storage medium for dispatching government affairs bills |
| CN116842165A (en) * | 2023-07-20 | 2023-10-03 | 中国银行股份有限公司 | Reason analysis methods, devices, equipment and media for negative customer service reviews |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111177389A (en) | NLP technology-based classification method, system and storage medium for power charge notification and customer appeal collection | |
| CN109063945A (en) | A kind of 360 degree of customer portrait construction methods of sale of electricity company based on Value accounting system | |
| CN108665184A (en) | A kind of power customer credit assessment method based on big data reference | |
| CN106780140A (en) | Electric power credit assessment method based on big data | |
| CN107507038A (en) | A kind of electricity charge sensitive users analysis method based on stacking and bagging algorithms | |
| CN102081781A (en) | Finance modeling optimization method based on information self-circulation | |
| CN111626543A (en) | Method and device for processing power related data | |
| CN110245879A (en) | A kind of risk rating method and device | |
| CN112396437A (en) | Trade contract verification method and device based on knowledge graph | |
| Wang et al. | Critical risk factors in PPP waste-to-energy incineration projects | |
| CN116563006A (en) | Business risk early warning method, equipment, storage medium and device | |
| CN119205305A (en) | A financial product intelligent matching and pre-credit method and system | |
| CN120317999A (en) | Account reporting method, device and electronic device based on OCR and Transformer decoder | |
| CN109558486A (en) | Electric power customer service client's demand intelligent identification Method | |
| CN116777592A (en) | Financial business approval method, device, equipment and storage medium | |
| CN116226293A (en) | A method and system for generation and management of power customer portraits | |
| CN117764692A (en) | Method for predicting credit risk default probability | |
| CN120634754A (en) | A multi-agent-based accounting business processing method and system | |
| CN107506899A (en) | A kind of photovoltaic apparatus evaluation method based on Life cycle cost | |
| CN112711618A (en) | Client appeal emotion analysis method and device | |
| CN112463893A (en) | Intelligent analysis system and method for network fund | |
| CN112308608B (en) | Data processing method, device, electronic equipment and storage medium for substation construction cost | |
| CN115099498A (en) | Customer demand prediction classification method and device | |
| CN107832408A (en) | It is a kind of that method is recommended based on the grounding grid defect of data label and entropy assessment | |
| Chen et al. | Research on sensitivity and strategy service of customer electricity charge based on big data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200519 |
|
| RJ01 | Rejection of invention patent application after publication |