CN113379432B - Sales system customer matching method based on machine learning - Google Patents
Sales system customer matching method based on machine learning Download PDFInfo
- Publication number
- CN113379432B CN113379432B CN202110748337.0A CN202110748337A CN113379432B CN 113379432 B CN113379432 B CN 113379432B CN 202110748337 A CN202110748337 A CN 202110748337A CN 113379432 B CN113379432 B CN 113379432B
- Authority
- CN
- China
- Prior art keywords
- model
- customer
- sales
- samples
- customer information
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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
- G06Q30/015—Providing customer assistance, e.g. assisting a customer within a business location or via helpdesk
- G06Q30/016—After-sales
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Marketing (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Business, Economics & Management (AREA)
- Finance (AREA)
- Economics (AREA)
- Development Economics (AREA)
- Artificial Intelligence (AREA)
- Strategic Management (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Accounting & Taxation (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a sales system customer matching method based on machine learning, which is characterized by comprising the following steps of firstly, carrying out data exploration on customer information in a sales system to obtain customer information; step two, performing data preprocessing on the customer information to obtain a training set and a test set; inputting character string features in the training set and the test set into an LDA topic model for topic classification; designing a sample for training the model; step five, a first model and a second model are built in a GBDT + LR construction mode, the first model predicts the transaction possibility of the customer, and the second model predicts the matching probability of the sales and the customer; and step six, sequentially inputting the information of the customer to be tested into the first model and the second model, and outputting the matching probability of the customer and each sale. The method achieves the effects of solving a plurality of invalid allocations caused by the random allocation of sales clues by the sales system and improving the allocation efficiency of customers.
Description
Technical Field
The invention relates to the field of customer distribution systems in sales management systems, in particular to a machine learning-based sales system customer matching method.
Background
For sellers in foreign trade industry, this kind of sellers basically adopt the way of telephone sales to do initial mining of customers. But there is a problem in that excessive development or ineffective development often occurs in the course of its telemarketing developing client, thereby causing a rise in company cost. For example, the current sales system can not actively and intelligently allocate customers according to actual business requirements of foreign trade industry, and the sales basically adopts a method of searching for customer attribute keywords to search customers and add target customers into a customer pool of the sales system. Therefore, a fast, efficient and targeted customer search recommendation system is needed to solve the cost problem of sales development.
Disclosure of Invention
The technical problem to be solved by the invention is to overcome the defects of the prior art and provide a sales system customer matching method based on machine learning.
In order to solve the technical problem, the invention provides a sales system customer matching method based on machine learning, which is characterized by comprising the following steps:
step one, carrying out data exploration on customer information in a sales system to obtain the customer information;
step two, data preprocessing is carried out on the customer information to obtain a training set and a testing set;
inputting character string features in the training set and the test set into an LDA topic model for topic classification;
designing a sample, wherein the sample is used for training the model;
step five, a first model and a second model are constructed in a GBDT + LR construction mode, the first model and the second model are trained by using samples, the first model predicts the transaction possibility of customers according to customer information in the samples, and the second model predicts the matching probability of sales and customers according to the customer information in the samples and the transaction possibility output by the first model;
and step six, sequentially inputting the information of the customer to be tested into the first model and the second model, and outputting the matching probability of the customer and each sale.
The data exploration in the first step is specifically as follows: extracting customer information from a sales system, wherein the attribute characteristics in the customer information comprise: customer ID, sales ID, contact record, purchase record, telephone, fax, customer source, extent of business, size of company staff, company registered capital, extent of business, company type, location, possibility of customer deal.
In the second step, the data preprocessing includes: deleting missing character strings in the attribute features, carrying out discrete feature conversion on the attribute features without substantial significance, converting the result into 0,1 labels, carrying out label statistics on customer sources and company types and converting the labels into virtual variables, carrying out data exploration on the scale of company personnel and the registered capital of companies, dividing data intervals, converting the virtual variables into the data intervals, and dividing customer information into a training set and a testing set by using the train _ test _ split function of skleann.
In the third step, quantitative parameter adjustment is carried out on the LDA theme model through the perfect calculation and chi-square test, the number of words used for the quantitative parameter adjustment is 7500 words, and the number of themes is 30.
In the fourth step, the samples in the sample design include corresponding first positive and negative samples, and corresponding second positive and negative samples, where the first positive sample is the customer information of a customer with a sales record, and the first negative sample is the customer information of a customer without a sales record; the second positive sample is customer information of a contact record and a purchase record with a sales ID in a preset time range; the second negative example is customer information with a contact record of the sales ID but no purchase record within a preset time frame.
In the fifth step, the first model includes a first GBDT feature selection model and a first LR model, the second model includes a second GBDT feature selection model and a second LR model, the first GBDT feature selection model and the second GBDT feature selection model both use a gradientboosting classifier under a sklern machine learning framework to perform data feature selection training, the output result is subjected to one-hot encoding processing by using OneHotEncoder, the processed vector and sample are trained by using logistic regression in sklern, the first GBDT feature selection model uses first positive and negative samples, and the second GBDT feature selection model uses second positive and negative samples.
In the second step, the data preprocessing further includes: the province names mixed in Chinese and English are processed into a Chinese and English format, units of different currencies in the registered capital of the unified company are unified, and the units are converted into data taking the Ten thousand yuan RMB as units through real-time exchange rate; the attribute features without substantive significance include: telephone, fax, no telephone or fax is 0, otherwise is 1.
In the fifth step, using GridSearch of skleran to carry out global parameter adjustment on the first GBDT feature selection model and the second GBDT feature selection model to obtain random _ state: 10. 0.1 for learning _ rate, 100 for n _ estimators, 5 for max _ depth, 300 for min _ samples _ split, 20 for min _ samples _ leaf, 301 for max _ features, 1.0 for subsample, and random _ state: 10. second GBDT feature selection models of left _ rate 0.1, n _ estimators 200, max _ depth 5, min _ samples _ split 500, min _ samples _ leaf 80, subsample 1.0.
The invention has the advantages that the matching is carried out based on the client information characteristics, the bargaining possibility and the preference of the sales development client, and the algorithm of GBDT + LR is used for calculating the matching percentage. In reality, the number of customers and the data volume matched with sales are very large, and the data is mostly a sparse matrix in a plurality of customer information characteristics and sales development preferences, so that the GBDT is very convenient, fast and accurate to use as characteristic processing, and LR as a final classification algorithm has very high calculation speed and good interpretability, so that a plurality of invalid distributions caused by random distribution of sales clues by a sales system are solved, and the customer distribution efficiency is improved.
Drawings
FIG. 1 is a simplified process flow diagram of an exemplary embodiment of the present invention;
FIG. 2 is a data diagram without virtual variable translation in an exemplary embodiment of the invention;
fig. 3 is a data diagram after virtual variable translation and interval processing in an exemplary embodiment of the invention.
Detailed Description
Machine learning is a discipline covering probability theory, statistics, and complex algorithms. The computer is used as a tool for machine learning in a big data environment, and the main purpose is to research how to effectively utilize information and extract hidden, effective and regular information from mass data. The very important core algorithm in the invention uses a Gradient Boosting Decision Tree (GBDT) in an integrated learning theory and a very classical classification algorithm Logistic Regression (LR).
GBDT is an integrated learning model, and each calculation is to reduce the residual error of the previous time, and to eliminate the residual error, we can model the gradient direction of the reduced residual error. Therefore, the key of the GRDT is to use the value of the negative gradient direction of the loss function in the current model as an approximate value of the residual error, and further fit a CART regression tree. Therefore, the algorithm determines that the algorithm can flexibly process various types of data, has high prediction accuracy under relatively less parameter adjusting time and is very suitable for fitting nonlinear data. LR is a log-linear model used to solve the binary problem, which outputs not the exact class, but a probability. The main advantage of LR is that the model is simple and easy to interpret.
The GBDT and LR algorithms, the combined algorithm proposed by Facebook in 2014, are mainly used for predicting advertisement click-through rate. However, there are some non-linear and some linear relationships between sales and customers, so that prediction using such algorithms must be configured specifically. The model automatically screens and combines features by using GBDT to generate a new discrete feature vector, and then the feature vector is used as an input feature of an LR model to generate a final prediction result.
The invention will be further described with reference to the drawings and exemplary embodiments:
a machine learning based sales system customer matching method as shown in fig. 1, comprising:
step 2, carrying out data preprocessing on the client information to obtain a training set and a test set;
step 4, designing a sample, wherein the sample is used for training a model;
and 6, sequentially inputting the information of the customer to be tested into the first model and the second model, and outputting the matching probability of the customer and each sale in the customer information.
The method specifically comprises the following steps:
step 2, data preprocessing is carried out on the customer information, and the data preprocessing comprises the following steps: performing data preprocessing on the client attribute characteristics obtained in the step one, wherein the data preprocessing comprises the following steps: processing province names with Chinese and English mixed into a Chinese and English format, deleting missing character strings in client attribute characteristics, unifying units of different currencies in company registered capital, and converting the units into data with the Ten thousand yuan RMB as a unit through real-time exchange rate; the cleaning conversion comprises discrete feature conversion of the data features without substantial significance, and the conversion of the result into a 0,1 label, wherein the data features without substantial significance comprise: the telephone, the fax machine,such asIf no telephone or fax is 0, otherwise, the number is 1, label statistics is carried out on the source of a client and the company type, the label statistics is carried out and converted into virtual variables, data exploration is carried out on the staff size of the company and the processed company registered capital, proper data intervals are divided, then virtual variable conversion is carried out on the intervals, and client information is divided into a training set and a testing set by using the train _ test _ split function of sklern;
and 3, inputting the character string characteristics in the training set and the test set into an LDA topic model for topic classification. Because the business scope data of a company is not in a fixed format, the data cannot be cluttered, quantitative parameter adjustment can only be carried out on the LDA theme model through perplexity calculation and chi-square test, the number of words used for quantitative parameter adjustment is 7500 words, the number of themes is 30, 10000 words, 7500 words and 5000 words are selected for theme model processing, in the three cases, the number of themes topic is 5,10,15,20,25,30 and 35 for modeling respectively, and finally, the scores of perplexity and chi-square test are optimal under the condition that the number of themes is 30 under the 7500 words.
Step 4, designing positive and negative samples, wherein the samples are used for training the model, and a GBDT + LR construction mode is selected to construct a customer deal possibility model, which specifically comprises the following steps: in sample data used for training the first model, a positive sample is a client with a sales record in the historical database, a negative sample is a client without the sales record in the historical database, and the GradientBoosting Classifier under the sklern machine learning framework is used for carrying out data feature selection training.
And 5-1, using a GridSearch tool of sklern to carry out global parameter adjustment on random _ state, left _ rate, n _ estimators, max _ depth, min _ samples _ split, min _ samples _ leaf, max _ features and subsample parameters, and obtaining a first GBDT feature selection model. And finally obtaining random _ state: 10. 0.1 for learning _ rate, 100 for n _ estimators, 5 for max _ depth, 300 for min _ samples _ split, 20 for min _ samples _ leaf, 301 for max _ features, and 1.0 for subsample.
And 5-2, each leaf node output by the GBDT feature selection model is an input dimension of the LR model, the output result is subjected to one-hot coding processing by using an OneHotEncoder, the processed vector and positive and negative samples are trained by Logistic regression in sklern, and the final client transaction possibility model is obtained and used for outputting the transaction possibility of each client.
And 5-3, constructing a second model, wherein the second model is a matching model of a salesman and a customer, the second model is a GBDT + LR model, all samples selected by the second model are customers with sales records, and each piece of data comprises a sales id and a customer id. The sales record comprises a door record and a telephone record. And a record joining the private customer pool containing the sales ID and customer ID. The input customer attribute features include: sales ID, customer phone and fax, customer source, industry of ownership, extent of business staff, company registered capital, industry of ownership, company type, location of other historical customers under sales ID, city, business industry, and customer province, city, extent of business that is private to the sales ID and the customer ID. The positive sample designed in the second model is a sales ID and a customer ID which are connected and achieve purchasing behavior within a preset time range; such as that sales x contacted or privately owned customer y for a certain period of time and that within 30 days a record of the purchase occurred for customer y. The negative examples are sales ID and customer ID which are related but not in purchasing behavior within a preset time range, sales x is related or private to customer y within a certain time period, and customer y does not have purchasing behavior within 30 days. There is no purchase for customer y at sales x, but a record of the purchase occurs at sales z, this record being a negative sample for sales x and a positive sample for sales z. And after the data are prepared, performing data preprocessing by using the data processing method as the second step. The negative examples in this step must have corresponding sales contact records, one sales id for each client id, and here only for client ids.
And 5-4, processing the text data by using an LDA model, setting parameters of the LDA in the same step as the step three, and setting 7500 words and 30 theme numbers. And the selected feature items are firstly subjected to feature selection training by using a GradientBoosting Classifier under a sklern machine learning framework.
Step 5-5, performing GBDT parameter adjustment, and obtaining a random _ state: 10. second GBDT feature selection models of left _ rate 0.1, n _ estimators 200, max _ depth 5, min _ samples _ split 500, min _ samples _ leaf 80, subsample 1.0.
And 5-6, carrying out one-hot coding processing on the output result by using a vector output by the GradientBoosting Classifier and using a oneHotEncoder. And training the processed vector and positive and negative samples by using Logistic regression in sklern, obtaining a final sales and client matching possibility model, and outputting the matching probability of each sales and client.
And 6, sequentially inputting the information of the customer to be tested into the first model and the second model, and outputting the matching probability of the customer and each sale in the customer information.
The invention provides a sales system client matching method based on machine learning, which solves the problem of low sales lead development efficiency, and provides a model for conjecturing probability based on sales preference and client attribute in the prior art, for example, referring to sales record data and client information of a China manufacturing network MIC sales management system in the embodiment, a GBDT + LR model is used for fitting the transaction possibility of all clients, and then the transaction possibility and some sales characteristics are used for fitting the matching degree of sales to the clients.
The invention is a nesting of two GBDT + LR models, the first GBDT + LR model is a client transaction model, the client data is solved through GBDT, the child node characteristics of the tree model are generated, and then the LR model is used for calculating the client transaction probability;
the samples taken by the second GBDT + LR model are all sales record samples, both singled and singled. And inputting the preference of the sales contact client, such as province record of the sales contact client, the classification of the industry theme, province and industry theme preference of the sales private client, and the characteristics of the contact client and the client transaction possibility calculated by the last model. GBDT is also used to process these characteristics, and finally the child node characteristics are output, and then LR is used to sell the corresponding client deal possibility output.
For a specific embodiment of the process with the Chinese and English mixed province names, for example, provinces in a small amount of customer information are Jiangsu or Jiangsu, so that Jiangsu needs to be unified. The reason is that such data is mistakenly recognized when the virtual variables of the feature engineering are processed, and therefore, the data needs to be unified into the same data form. The specific examples of unification of different monetary units in the registered capital in step two are, for example, that some data have registered capital of 1.6 million dollars, or 160 ten thousand dollars, and most data units are million RMB, so that the data need to be unified into 1072 million RMB. The processing is also beneficial to the interval conversion in the subsequent characteristic engineering and then the conversion into the virtual variable.
For the example of the conversion of the virtual variable, for example, two variables such as province and registered capital, the data without the conversion of the virtual variable is shown in FIG. 2. The data after the virtual variable transformation and interval processing is shown in fig. 3. Data and intervals in the example are not real data for privacy reasons.
The explanation for joining the private customer pool is that the private customer privatizes the record, and all other sales of the private customer cannot be inquired from the sales operating system until the sales and the customer release the private relationship.
For the explanation of the preferred features, model one calculates a generic deal probability for this customer id across MIC sales in terms of only the customer's own attributes (customer phone and fax, customer source, industry of business, business scope, company staff size, company registered capital, industry of business, company type, location), but model two aims at highlighting the match. There is also a need to incorporate sales characteristics for each sale, such as customer preferences previously sold or contacted, e.g. (locations of historical customers, cities, business industries, and provinces of each customer ID that are historical private under the sales ID, cities, business boundaries), which are sales behavior preference characteristics under the sales ID. And the probability of a deal of the model one is only the probability of a common deal under one customer id, and on the basis of the probability of a deal, namely the matching probability, of the customer id under the sales id needs to be calculated according to the sales preference characteristics of each sales id.
The invention relates to a sales system customer matching method based on machine learning, which is used for matching based on customer information characteristics, transaction possibility and the preference of sales development customers and calculating the matching percentage by using a GBDT + LR algorithm. In reality, the number of clients and the data volume matched with sales are very large, and most data present a sparse matrix in a plurality of client information characteristics and sales development preferences, so that the adoption of GBDT as characteristic processing is very convenient, fast and accurate, and LR as a final classification algorithm also has very high calculation speed and good interpretability, so that a plurality of invalid allocations caused by a sales system when sales clues are randomly allocated are solved, and the client allocation efficiency is improved.
The above embodiments do not limit the present invention in any way, and all other modifications and applications that can be made to the above embodiments in equivalent ways are within the scope of the present invention.
Claims (6)
1. A sales system customer matching method based on machine learning is characterized by comprising the following steps:
step one, carrying out data exploration on customer information in a sales system to obtain the customer information;
step two, data preprocessing is carried out on the customer information to obtain a training set and a testing set;
inputting character string features in the training set and the test set into an LDA topic model for topic classification;
designing a sample, wherein the sample is used for training the model;
in the fourth step, the samples in the sample design include corresponding first positive and negative samples, and corresponding second positive and negative samples, where the first positive sample is customer information with a sales record, and the first negative sample is customer information without a sales record; the second positive sample is customer information of a contact record and a purchase record with a sales ID in a preset time range; the second negative sample is the customer information with the contact record of the sales ID but no purchase record in the preset time range;
step five, a first model and a second model are constructed in a GBDT + LR construction mode, the first model and the second model are trained by using samples, the first model predicts the transaction possibility of the customer according to the customer information in the samples, and the second model predicts the matching probability of the sales and the customer according to the customer information in the samples and the transaction possibility output by the first model;
in the fifth step, the first model includes a first GBDT feature selection model and a first LR model, the second model includes a second GBDT feature selection model and a second LR model, the first GBDT feature selection model and the second GBDT feature selection model both use a gradientboosting classifier under a sklern machine learning framework to perform data feature selection training, the output result is subjected to one-hot encoding processing by using onehotencor, the processed vector and sample are trained by using logistic regression in sklern, the first GBDT feature selection model uses first positive and negative samples, and the second GBDT feature selection model uses second positive and negative samples;
and step six, sequentially inputting the information of the customer to be tested into the first model and the second model, and outputting the matching probability of the customer and each sale.
2. The machine learning-based sales system customer matching method according to claim 1, wherein the data exploration in the first step is specifically as follows: extracting customer information from a sales system, wherein the attribute characteristics in the customer information comprise: customer ID, sales ID, contact record, purchase record, telephone, fax, customer source, scope of business, size of company personnel, company registered capital, company type, location, possibility of customer deal.
3. The machine learning-based sales system customer matching method of claim 2, wherein in the second step, the data preprocessing comprises: deleting missing character strings in the attribute features, performing discrete feature conversion on the attribute features without substantial significance, converting results into 0,1 labels, performing label statistics on customer sources and company types and converting the labels into virtual variables, performing data exploration on the scale of company personnel and company registered capital, dividing data intervals, converting the virtual variables into the data intervals, and dividing customer information into a training set and a test set by using a train _ test _ split function of skearn.
4. The machine learning-based sales system customer matching method according to claim 3, wherein in the third step, quantitative parameter adjustment is performed on the LDA topic model through perplexity calculation and chi-square test, the quantitative parameter adjustment has a word number of 7500 words and a topic number of 30 words.
5. The machine learning-based sales system customer matching method of claim 4, wherein in the second step, the data preprocessing further comprises: the province names mixed in Chinese and English are processed into a Chinese and English format, units of different currencies in the registered capital of the unified company are unified, and the units are converted into data taking the Ten thousand yuan RMB as units through real-time exchange rate; the attribute features without substantive significance include: telephone, fax, no telephone or fax is 0, otherwise is 1.
6. The machine learning-based sales system customer matching method of claim 5, wherein in the fifth step, the first GBDT feature selection model and the second GBDT feature selection model are globally parameterized using GridSearch of skleann to obtain random _ state: 10. 0.1 for learning _ rate, 100 for n _ estimators, 5 for max _ depth, 300 for min _ samples _ split, 20 for min _ samples _ leaf, 301 for max _ features, 1.0 for subsample, and random _ state: 10. second GBDT feature selection models of left _ rate 0.1, n _ estimators 200, max _ depth 5, min _ samples _ split 500, min _ samples _ leaf 80, subsample 1.0.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110748337.0A CN113379432B (en) | 2021-06-30 | 2021-06-30 | Sales system customer matching method based on machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110748337.0A CN113379432B (en) | 2021-06-30 | 2021-06-30 | Sales system customer matching method based on machine learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113379432A CN113379432A (en) | 2021-09-10 |
| CN113379432B true CN113379432B (en) | 2022-07-19 |
Family
ID=77580599
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110748337.0A Active CN113379432B (en) | 2021-06-30 | 2021-06-30 | Sales system customer matching method based on machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113379432B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116629796A (en) * | 2023-05-26 | 2023-08-22 | 深圳科海数信科技有限公司 | Full-period sales management system based on artificial intelligence and big data |
| CN116580249B (en) * | 2023-06-06 | 2024-02-20 | 河北中废通拍卖有限公司 | Method, system and storage medium for classifying beats based on ensemble learning model |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112561568A (en) * | 2020-12-04 | 2021-03-26 | 中国农业银行股份有限公司深圳市分行 | Target customer prediction method, device and storage medium |
| CN113052705A (en) * | 2021-04-13 | 2021-06-29 | 粤港澳国际供应链(广州)有限公司 | Supply chain financial service process management method and device |
-
2021
- 2021-06-30 CN CN202110748337.0A patent/CN113379432B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112561568A (en) * | 2020-12-04 | 2021-03-26 | 中国农业银行股份有限公司深圳市分行 | Target customer prediction method, device and storage medium |
| CN113052705A (en) * | 2021-04-13 | 2021-06-29 | 粤港澳国际供应链(广州)有限公司 | Supply chain financial service process management method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113379432A (en) | 2021-09-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110825882B (en) | Knowledge graph-based information system management method | |
| EP3144822A1 (en) | Tagging text snippets | |
| CN111914096A (en) | Public transport passenger satisfaction evaluation method and system based on public opinion knowledge graph | |
| CN107679234A (en) | Customer service information providing method, device, electronic equipment, storage medium | |
| CN110245240A (en) | A kind of determination method and device of problem data answer | |
| CN104933081A (en) | Search suggestion providing method and apparatus | |
| CN113379432B (en) | Sales system customer matching method based on machine learning | |
| US11804216B2 (en) | Generating training datasets for a supervised learning topic model from outputs of a discovery topic model | |
| WO2023035330A1 (en) | Long text event extraction method and apparatus, and computer device and storage medium | |
| CN113505242A (en) | Method and system for automatically embedding knowledge graph | |
| CN115600605A (en) | Method, system, equipment and storage medium for jointly extracting Chinese entity relationship | |
| CN113553431B (en) | User tag extraction method, device, equipment and medium | |
| US11966698B2 (en) | System and method for automatically tagging customer messages using artificial intelligence models | |
| CN111626331B (en) | Automatic industry classification device and working method thereof | |
| CN112749238A (en) | Search ranking method and device, electronic equipment and computer-readable storage medium | |
| Krishnareddy et al. | AI-based Fuzzy Clustering System for Improving Customer Relationship Management | |
| CN115841365A (en) | Model selection and quotation method, system, equipment and medium based on natural language processing | |
| KR20230059364A (en) | Public opinion poll system using language model and method thereof | |
| CN110851572A (en) | Session labeling method and device, storage medium and electronic equipment | |
| Spada et al. | WHAT USERS WANT: A NATURAL LANGUAGE PROCESSING APPROACH TO DISCOVER USERS'NEEDS FROM ONLINE REVIEWS | |
| CN116468460B (en) | Consumer finance customer image recognition system and method based on artificial intelligence | |
| Bharadi | Sentiment Analysis of Twitter Data Using Named Entity Recognition | |
| Chen et al. | Towards accurate search for e-commerce in steel industry: a knowledge-graph-based approach | |
| CN111444714B (en) | Text analysis device and method and model training method | |
| CN117852553B (en) | Language processing system for extracting component transaction scene information based on chat record |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |