Intelligent connection touch method and system based on big data technology

Abstract

The invention relates to an intelligent connection touch method and system based on a big data technology, and belongs to the technical field of computers, authorized compliant user identification requirements are encrypted and transmitted to a platform, the platform establishes a user model through a big data model algorithm according to user identification, and intelligent connection touch steps are completed according to the user model in combination with touch capability. The intelligent connection arriving step further comprises an intelligent connection verification step, a lost connection repairing step and an intelligent arriving step. The method gives full play to the advantages of operator data, constructs the user screening model based on the data of each dimension of the user, carries out the secondary processing and deep analysis of the data, screens the potential high-willingness users, complements the user portraits, enables the diversified touch, and improves the customer obtaining efficiency.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an intelligent connection method and system based on big data technology. Background Art

[0002] With the rapid development of information technology, the role of individuals in the operation of modern enterprises has become increasingly important. However, in this era of endless communication methods, the phenomenon of being unable to maintain normal contact with the outside world has become increasingly common, causing great troubles in many fields such as credit assessment of financial institutions, advancement of judicial procedures, and efficient operation of government management. In the face of the growing number of dishonest people, it is urgent to establish a comprehensive and efficient loss of contact repair mechanism to ensure the accuracy and completeness of information, and then restore contact with these lost persons.

[0003] In actual solutions, the challenges and difficulties faced by the missing contact repair technology should not be underestimated. First, information collection is difficult. Due to the diversity and complexity of data sources, existing technologies often find it difficult to cover all possible sources of information, resulting in the omission of key information. This not only affects the accuracy of subsequent work, but also may miss the opportunity to reconnect with the missing person. Secondly, technical problems also need to be solved urgently. In the process of screening, verifying and repairing information, existing technical means often have problems such as slow real-time processing and low security. This not only affects the efficiency of the repair work, but also may bring potential risks to data security. How to improve the efficiency of the repair work while ensuring data security has become a major problem faced by the missing contact repair process. In addition, legal and ethical considerations cannot be ignored. In the process of missing contact repair, we must not only comply with relevant laws and regulations, but also respect personal privacy and data security. How to effectively use the personal information of the missing person within the scope permitted by law, carry out necessary repair work, and avoid risks such as data leakage and abuse is a problem we must face in practice.

[0004] In summary, the existing lost connection repair technology lacks automated and intelligent repair tools and methods, making it impossible for the repair process to respond quickly and adapt to changes. When dealing with a large number of lost users, existing technologies are often unable to respond effectively, resulting in low repair efficiency and poor results, resulting in low repair rate and reach rate. When reaching users, due to limitations in information transmission channels or problems with user receiving devices, the reach rate is often far lower than expected. At the same time, the lack of accurate user portraits and personalized reach strategies greatly reduces the reach effect. For lost users who have been successfully repaired, SMS is often used for contact, and the form is relatively simple. In addition, in the process of repair and contact, a large amount of user data is involved in the transmission and processing. However, existing technologies often do not pay enough attention to data security and privacy protection, and there is a risk of data leakage and abuse. This will not only affect user trust, but may also cause legal disputes and compliance issues. Summary of the invention

[0005] The present invention intends to provide an intelligent connection method and system based on big data technology to solve the deficiencies in the prior art. The technical problems to be solved by the present invention are achieved through the following technical solutions.

[0006] A first aspect of the present invention provides an intelligent connection and contact method based on big data technology, the method comprising: transmitting the authorized and compliant user identification requirements to the platform through encryption, the platform establishing a user model based on the user identification through a big data model algorithm, and completing the intelligent connection and contact step based on the user model combined with the contact capability, the intelligent connection and contact step further comprising an intelligent connection verification step, a loss of connection repair step and an intelligent arrival step; In the Zhilian verification step, the customer transmits the irreversibly encrypted user identification through the interface, matches the existing Zhilian verification model according to the transmitted user identification information, and returns the verification result to the user. The acceptance result is to determine whether the user is connectable. If not, repair is recommended. If connectable, the degree of connectability is further determined to be high, medium or low. The lost contact repair step constructs a repair model based on multi-dimensional data. The repair model determines that the contact information to be repaired is the phone number currently on the network, and removes the recently inactive contact information to ensure the reach rate. At the same time, it determines the user's complaint tendency and controls the complaint rate after the reach. The repair model outputs the user's contact information; in the reach stage after the repair, the China Unicom privacy number capability is used to protect the user's privacy and the real number is not returned; The intelligent reaching step builds user screening models for different industries based on the user's basic information, behavioral data, and consumption data in combination with different scenarios, performs secondary processing and in-depth analysis on the data, screens potential high-willing users, completes user portraits, and enables diversified reach.

[0007] In combination with the first aspect, the above-mentioned intelligent connection verification model verifies whether the mobile phone number belongs to the user himself based on different dimensional data of the user's network access date, online status, activity in the past three months, online time and secondary card verification, and uses a logistic regression algorithm to construct the user's connection score and repair score. The connection score is to score whether the user is reachable, and the connection is divided according to the score and the result is returned; the connection is returned according to the connection score. If it is not connectable, whether to recommend repair is returned based on whether there is an online contact method under the name, and the repair score returns whether to recommend repair.

[0008] Combined with the first aspect, the above-mentioned Zhilian verification model includes data selection, data processing, IV calculation and feature selection, logistic regression algorithm and parameter tuning, verification set evaluation and output scoring model; Data selection: select user data for a single month, classify 34 feature fields and 1 label field based on user samples, and divide the data into 8:2. 80% of the data will be used for training the connectable model, and the remaining 20% ​​of the data will be used for performance testing of the connectable model and reference setting of the scoring threshold, so as to examine the generalization ability of the model and the rationality of the connectable parameter setting; Data processing further includes missing value filling, feature binning and WOE binning. Random forest algorithm is used to fill missing values. Random forest classifier is used to fill missing values ​​when the field value type is discrete, and random forest regressor is used to fill missing values ​​when the field value type is continuous. Chi-square binning method is used to perform feature binning on continuous variables to achieve data discretization. WOE binning is performed by calculating the woe value for each bin or each value of the discrete variable using the following formula: (1) Bad total Indicates the total number of bad samples, Bad X=Xi represents bad samples and X is the number of the i-th group, Good X=Xi Good represents good samples and X is the number of the i-th group; total Indicates the total number of good samples; IV calculation and feature selection, variable screening is done through the linear correlation between features and the IV value of features: the stronger the correlation between features, the stronger the feature collinearity will be, and the greater the impact on label prediction will be. Before modeling, try to remove redundant features, that is, features that have strong correlation with other features but little contribution to the label; the IV value can be understood as the weighted average of woe. The larger the IV value, the greater the contribution of the feature to the label amount; finally, the modeling parameters and features are determined based on the IV value and feature selection; Logistic regression algorithm training and parameter tuning: Use the feature data of the input model to train the logistic regression algorithm, set the logistic regression algorithm parameter search grid, use the area under the AUC (Area Under Curve) curve and the coordinate axis as the algorithm performance reference, use the random search algorithm to optimize the algorithm parameters and obtain the optimal parameter set; the AUC value ranges from 0 to 1, and the larger the value, the better the model classification performance; Validation set evaluation: Use a validation set accounting for 20% to evaluate the model effect. The model evaluation indicators include accuracy, recall, precision, F1 value, AUC, and KS value. The model evaluation of logistic regression generally uses AUC to calculate the AUC of training samples and test samples respectively, and then evaluates whether the model can be put into production based on AUC; Output the scoring model, convert the model into a score output, and divide the connectability into high and low according to the score. Do not use probability to convert the score, but use the linear part to convert it into the score. The final result is the connectability of the input user.

[0009] Combined with the first aspect, the WOE binning further includes two bins for the number of active times: [5000, 10000) and [10000, 30000). The woe of the former bin is 2, and the woe of the latter bin is -1, indicating that the possibility of users active in [10000, 30000) being unreachable is low, and the possibility of users in [5000, 10000) being unreachable is high.

[0010] In combination with the first aspect, the model is evaluated according to AUC, further including: AUC value > 0.63: the model is distinguishable on whether it can be connected, but the distinguishability is not enough to meet the verification scenario requirements of intelligent connection verification and cannot be put into production; AUC value > 0.73: the model is excellent and can be put into production; AUC value: 0.8029, discrimination evaluation index KS: 0.4500, accuracy: 0.7954 meet the production requirements.

[0011] In combination with the first aspect, the above-mentioned intelligent arrival step further includes: Establish a data screening rule model, which deletes invalid users and non-mobile network users, and only retains the mobile network and online users that are currently in normal use; set a network access time threshold to exclude users who have been on the network for less than six months from the promotion candidate list, aiming to focus promotion resources more on stable users who have established certain usage habits; introduce a "three-no" user identification algorithm to identify users who have no recent call, SMS or data traffic usage records, and remove them from the promotion set; A user screening model is established to screen out users who meet the requirements. The user screening model first performs fine division based on the user's age information, and only retains the user group between the ages of 22 and 45. Then, the user screening model screens out users who have recently visited the relevant website more than 1 time and less than 8 times, and sorts them from the recent to the longest visit time and the low to high visit times to obtain a user result set. Obtain a user screening model for operator product cloud phones, and screen users who have demand and willingness for cloud phone operator product projects based on user interaction information and the user's own attribute characteristics; perform personalized processing based on the unique characteristics of the target users of the product cloud phone.

[0012] In combination with the first aspect, the above-mentioned user screening model for operator product cloud phones further includes: for users who frequently participate in activities, in-depth mining of user behavior data in the past month, comprehensive consideration of the frequency of users' clicks and stay time on a certain type of application, and accurate screening of high-frequency user groups based on intelligent sorting algorithms to form a target user set; for the target group of college students, setting an age threshold, limiting the user age to under 30 years old as a preliminary screening condition, using geographic location information, combined with electronic fence technology, to determine whether the user is located in the campus area as a key basis for identifying students on campus, integrating traffic usage and learning application usage habits, and through comprehensive evaluation, ultimately outputting highly accurate target results for college students.

[0013] A second aspect of the present invention provides an intelligent connection and contact system based on big data technology, the system comprising: a platform accepts encrypted and transmitted authorized and compliant user identification requirements, the platform establishes a user model based on the user identification through a big data model algorithm, and completes an intelligent connection and contact module based on the user model combined with the contact capability, the intelligent connection and contact module further comprising an intelligent connection verification submodule, a lost connection repair submodule and an intelligent arrival submodule; The Zhilian verification submodule matches the existing Zhilian verification model according to the irreversibly encrypted user identification information transmitted by the customer, and returns the verification result to the user. The acceptance result is to determine whether the user is connectable. If not, repair is recommended. If connectable, the degree of connectability is further determined to be high, medium or low. The lost contact repair submodule builds a repair model based on multi-dimensional data. The repair model determines that the contact information to be repaired is the phone number currently on the network, and removes the recently inactive contact information to ensure the reach rate. At the same time, it determines the user's complaint tendency and controls the complaint rate after the reach. The repair model outputs the user's contact information; in the reach stage after the repair, the China Unicom privacy number capability is used to protect the user's privacy and the real number is not returned; The intelligent reach submodule builds user screening models for different industries based on the user's basic information, behavioral data and consumption data, combined with different scenarios, and performs secondary processing and in-depth analysis on the data to screen potential high-willing users, complete user portraits, and enable diversified reach.

[0014] The third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of an intelligent connection and contact method based on big data technology are implemented.

[0015] The fourth aspect of the present invention provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of an intelligent contact method based on big data technology. The present invention uses data analysis to confirm the connectability of the contact information of the existing lost user group through the early intelligent connection verification process, reduces the cost of manual invalid contact, and uses data to enable the entire process of lost connection repair. In addition, in the face of the situation where users of banks, securities, and government customers have opened accounts but have not used them for a long time, a dormant wake-up sub-service is provided to reach users. Finally, a variety of intelligent contact modes are provided, and users are reached through various methods such as email, SMS, voice, mailbox, 5G message number, and corporate business card, enriching the contact mode. The present invention also gives full play to the advantages of operator data, builds a user screening model based on user data of various dimensions, performs secondary processing and in-depth analysis of the data, screens potential high-willing users, completes user portraits, enables diversified contact, and improves customer acquisition efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a smart connection and reaching technology solution based on big data technology of the present invention; Figure 2 It is a flow chart of the intelligent connection method based on big data technology of the present invention. DETAILED DESCRIPTION

[0017] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0018] It should be noted that the above detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the art to which the present application belongs.

[0019] The present invention provides a technical solution based on the full process integration of "data model + loss repair + reach capability", which is committed to creating an industry data model customized for operator data resources, integrating intelligent connection verification and intelligent customer service, 5G messaging and other reach capabilities, and using the B domain, O domain, and M domain data in the operator's personal user data and the business party's operation data to perform data modeling, and return the results to the customer, providing three sub-services including intelligent connection verification, loss repair and intelligent reach to industries such as banking, consumer finance, insurance, automobiles, and governments. By performing secondary processing and in-depth analysis of data and enriching the reach methods, the full process service capabilities and efficiency are improved.

[0020] like Figure 1 As shown, the present invention relies on China Unicom's online credit reporting product service platform. Under the premise of complying with national laws, regulations and regulatory policies and ensuring information security and user privacy, the customer will transmit the authorized and compliant user identification requirements to the platform through encryption. The platform will establish a user model based on the user identification through the big data model algorithm. And according to the user model combined with the reach capability, a targeted intelligent connection and reach solution is completed. The intelligent connection and reach solution mainly consists of three parts: intelligent connection verification, loss of connection repair and intelligent reach, covering multiple scenarios of user reach, and solving the problems of inability to find users, low user reach efficiency, and inaccurate user reach while ensuring user privacy and security.

[0021] Data modeling is performed based on the B-domain, O-domain, and M-domain data in the operator's personal user data and the business party's operational data, and the results are returned to the customer. Three sub-services including intelligent connection verification, lost connection repair, and intelligent arrival are provided to industries such as banking, consumer finance, insurance, automobiles, and government.

[0022] like Figure 2 As shown in the figure, the intelligent connection and reach solution mainly consists of four parts: intelligent connection verification, loss connection repair and intelligent arrival.

[0023] Smart Verification Intelligent connection verification is mainly implemented through platform interfaces to meet the needs of Internet, financial and other customers for maintaining user stock. It makes comprehensive judgments on user effectiveness, connectivity, high complaint risks, etc. and returns the results.

[0024] The user transmits the irreversibly encrypted user ID through the interface, matches the existing Zhilian verification model according to the transmitted user ID information, and returns the verification result to the user. The verification result mainly distinguishes whether the user is connectable. If not, it is recommended to repair. If it is connectable, it is further determined whether the connectability is high, medium or low. The core of Zhilian verification lies in the verification model. The model verifies whether the mobile phone number belongs to the user through different user dimensions of operator data, including but not limited to user network access date, network status, activity in the past three months, network duration, secondary card verification and other information, and then uses the logistic regression algorithm to build the user's connectability score and repair score. The connectability is returned according to the connectability score. If not, it returns whether it is recommended to repair according to whether there is an online contact method under the name. The repair score returns whether it is recommended to repair. The connectability score is to score whether the user is reachable, and then divide the connectability according to the score and return the result.

[0025] The construction process of Zhilian's verification model includes data selection, data processing (missing value filling, feature binning, WOE binning), IV calculation and feature selection, logistic regression algorithm and parameter tuning and validation set evaluation, and finally outputs the scoring model.

[0026] 1. Data selection We selected user data for a single month. The data generally includes 2,436,607 users (samples), 34 feature fields (7 discrete features, 27 continuous features), and 1 label field (values ​​are 0 and 1, 1 represents unlinkable, and the unlinkable ratio is about 3.06%). The data is divided into 8:2. 80% of the data will be used for training the linkable model, and the remaining 20% ​​of the data will be used for performance testing of the scoring card model and reference setting of the scoring threshold, so as to examine the generalization ability of the model and the rationality of the scoring card parameter setting.

[0027] To ensure the distribution consistency of the training set and the validation set, as shown in Table 1, the data is divided into layers according to the proportion of labels. After division: Dataset usage The label is 0 (connectable) The label is 1 (not connectable) train 1889595(96.94%) 59690(3.06%) verify 472399(96.94%) 14923(3.06%) Table 1 Data types and labels 2. Data processing (missing value filling, feature binning, WOE binning) The random forest algorithm is used to automatically fill in missing values. When the field value type of the missing value is discrete, the random forest classifier is used to fill it. When the field value type of the missing value is continuous, the random forest regressor is used to fill it.

[0028] The chi-square binning method is used to automatically bin continuous variables to achieve data discretization, such as converting age 23 years old into age in the interval [20,25).

[0029] For each bin or each value of the discrete variable, the Woe value is calculated using the following formula (1) for Woe encoding: (1) Bad total Indicates the total number of bad samples, Bad X=Xi represents bad samples and X is the number of the i-th group, Good X=Xi Good represents good samples and X is the number of the i-th group; total Indicates the total number of good samples; For example, if the number of active times has two bins [5000, 10000) and [10000, 30000), the woe of the former bin is 2, and the woe of the latter bin is -1, which means that the possibility of users active in [10000, 30000) being unreachable is small, and the possibility of users in [5000, 10000) being unreachable is high.

[0030] 3. IV calculation and feature selection Variable screening (feature selection) is performed through the linear correlation between features and the IV value of the features: the stronger the correlation between features (the threshold is set to 0.9), the stronger the feature collinearity will be, and the greater the impact on label prediction will be. Before modeling, try to remove redundant features (that is, features that are highly correlated with other features but have a small contribution to the label).

[0031] The IV value can be understood as the weighted average of woe. The larger the IV value, the greater the contribution of the feature to the label amount.

[0032] Finally, the mold parameters and features are determined based on the IV value and feature selection.

[0033] 4. Logistic regression algorithm training and parameter tuning The feature data of the input model is used to train the logistic regression algorithm, and the logistic regression algorithm parameter search grid is set. AUC (Area Under Curve) is the area under the curve and the coordinate axis. As a reference for algorithm performance, a random search algorithm is used to optimize the algorithm parameters to obtain the optimal parameter set; the AUC value ranges from 0 to 1. The larger the value, the better the model classification performance.

[0034] 5. Validation set evaluation A validation set accounting for 20% is used to evaluate the model effect. The model evaluation indicators include accuracy, recall, precision, F1 value, AUC, and the discrimination evaluation indicator KS value. The model evaluation of logistic regression generally uses AUC to calculate the AUC of training samples and test samples respectively, and then evaluates whether the model can be put into production based on AUC.

[0035] AUC>0.63: the model can distinguish whether it can be connected (cannot be put into production), AUC>0.68: it is effective (cannot be put into production), AUC>0.73: the model is excellent (can be put into production). According to the evaluation model: AUC: 0.8029, ks: 0.4500 accuarcy: 0.7954 meets the requirements for production.

[0036] 6. Output scoring model Outputting the scoring model means converting the model into a score output and dividing the connectability according to the score. Instead of using probability to convert the score, the linear part is used to convert the score. The advantage of this is that the sum of the scores of each feature x is the total score. If the probability is used to convert the score, a single feature increases by S points, but the total score does not increase by s points. The final result is the connectability of the input user.

[0037] Lost connection repair For financial, insurance, and government customers who have overdue loans, insurance renewals, and government notifications, and who cannot be reached through the reserved contact after verification, the lost contact person must be reached through the lost contact repair service to complete the effective delivery of the notification. At the same time, there are also cases where users of banking, securities, and government customers have opened accounts that have not been used for a long time. In accordance with the requirements of relevant regulatory authorities, reminders, cancellations, and information updates must be made as soon as possible. When the reserved contact method cannot be reached after verification, the user must be reached through the sleep wake-up service.

[0038] Based on the above different scenarios, it is necessary to repair the user's contact information and provide efficient means of contact.

[0039] In the repair phase, we use multi-dimensional data of users to make judgments and build a repair model. On the one hand, the repair model ensures that the contact information to be repaired is the current phone number of the user, and at the same time eliminates the recently inactive contact information to ensure the reach rate. On the other hand, we judge the user's complaint tendency and whether he is a red-list or black-list user to control the complaint rate after reaching out. The final model outputs the user's contact information.

[0040] During the post-repair contact phase, China Unicom's privacy number capabilities are used to protect user privacy and the real number will not be returned.

[0041] Specific steps in the repair phase: To ensure the privacy of personal data, all interactions use a unique ID. The customer uploads an encrypted ID number, and the repair model is used to find their valid contact information, and the customer is informed of the success or failure of the repair. When the repair is successful, the customer can bind the privacy number through the unique ID. The system binds the valid mobile phone number to the privacy number, and the lost user can be reached by calling the privacy number to complete the entire lost connection repair process. The repair connection can use the interface method, or the provided Sass service can be used to perform repair + contact operations on the interface. If the contact mostly uses fixed words and templates, AI robot outbound calls are also supported.

[0042] It also integrates robot outbound calling capabilities to provide an integrated solution from repair to outbound calling. After the call is made, call details can be queried and recordings can be obtained and quality checked.

[0043] Smart arrival In business scenarios such as activating existing users and attracting new users in the Internet, finance and other industries, we provide solutions to problems such as difficulty in accurately identifying target users, so as to accurately locate target users and improve marketing conversion effects.

[0044] Taking full advantage of the operator's data, we build a user screening model based on the user's data of various dimensions, conduct secondary processing and in-depth analysis of the data, screen potential high-willing users, complete user portraits, enable diversified reach, improve customer acquisition efficiency, and launch "smart reach" services. This project adheres to the principles of high flexibility and industry adaptability, and carefully builds a diversified user model system to accurately connect and meet the unique needs of cross-industry customers. Through in-depth analysis of the business logic and user behavior characteristics of different industries, we have successfully created customized data models. These models not only share a basic and efficient data screening framework to ensure the standardization and consistency of data processing, but also, on this basis, tailor a series of unique personalized screening rules for each industry's unique application scenarios. Here, taking the financial industry's small loan industry and the operator's cloud phone product marketing promotion as examples, we show our unified data screening framework and the modeling process of personalized user screening models for the above specific scenarios.

[0045] Data screening framework. First, in order to ensure that promotional information is effectively delivered to mobile phone numbers in normal use, the model deletes invalid users and non-mobile network users, and only retains mobile network users and users in normal use. In order to further enhance the experience of new users, the framework specifically sets a threshold for network access time, excluding users who have been on the network for less than six months from the promotion candidate list, aiming to focus promotion resources more on stable users who have established certain usage habits. In addition, the framework also introduces the "three-no user" identification logic, that is, to identify users who have no recent call, SMS or data traffic usage records and remove them from the promotion collection. This setting is intended to avoid ineffective promotion of potential inactive users and improve promotion efficiency. More importantly, the project fully respects the wishes of users. For users who clearly indicate that they do not accept SMS and telephone marketing services, especially those who have complained about such promotions, a strict exclusion mechanism is implemented to ensure that they are no longer included in the marketing user list. This practice not only maintains a good user experience, but also reflects the company's respect for user privacy and preferences. In summary, the above user screening conditions constitute a unified data screening framework, which is applied to the user modeling rules of each industry.

[0046] Financial micro-loan user screening model. In order to screen out users who are willing to use micro-loan products in the near future based on user interaction information and user's own attribute characteristics, targeted promotion and marketing are carried out. In addition to following the unified modeling specifications within the data screening framework, the model deeply combines the unique characteristics of micro-loan financial users and past interaction history, and implements a highly personalized processing strategy. Specifically, the model first makes a fine division based on the user's age information, retaining only the user group between 22 and 45 years old. This setting has two considerations: on the one hand, excluding users under 22 years old is mainly to avoid the financial loan risks that may be involved in minors and students, and to ensure business compliance and social responsibility; on the other hand, removing users over 45 years old is based on the statistical analysis results of the loan tendency of this age group, while taking into account the specific risks and complexities that may accompany loans for the elderly, so as to optimize the composition of the loan portfolio and risk management efficiency. In addition, in order to accurately screen users who are interested in and have needs for financial loan projects, the model screens out users who have visited financial loan apps or websites more than 1 time and less than 8 times in the past three days, and sorts them from recent to distant visit time and from low to high visit times to obtain a user result set.

[0047] Operator product cloud phone user screening model. In order to screen users with demand and willingness for cloud phone operator product projects based on user interaction information and user's own attribute characteristics for targeted promotion and marketing. In addition to following the unified modeling specifications within the data screening framework, the model also performs personalized processing based on the unique characteristics of the target users of the product cloud phone. For example, the target users of cloud phones are mainly college students, etc., so this project carefully designs a variety of model versions to accurately identify and serve these different categories of target users. Specifically, for users who frequently participate in activities, the model deeply mines user behavior data in the past month, and implements an intelligent sorting algorithm by comprehensively considering the user's click frequency and stay time on a certain type of application, so as to accurately screen out high-frequency user groups and form a target user set. For the target group of college students, the model adopts a more refined identification strategy. First, set the age threshold and limit the user age to under 30 years old as the initial screening condition; then, use geographic location information, combined with electronic fence technology, to determine whether the user is located in the campus area, which is used as the key basis for identifying students on campus. In addition, the model also incorporates other multi-dimensional rules, such as traffic usage, learning application usage habits, etc. Through comprehensive evaluation, it ultimately outputs a highly accurate target result set for college students.

[0048] The core of intelligent reach service lies in user screening model and various reach methods. Based on the user's basic information, behavior data, consumption data, etc., combined with different scenarios, user screening models for different industries are built. For the financial industry, focus on the user's consumption capacity and build a financial consumption model. With consumption data as the core, behavior data assists in determining whether the user is the target user of financial products. For the education industry, an education tendency model is built through the user's family attributes, app browsing behavior, etc., to determine the willingness of educational products. After screening the current users, the operator's reach capabilities can be used to efficiently and accurately reach users through voice, text messages, 5G messages, corporate business cards, emails and other methods.

[0049] The intelligent contact solution provides customers with a complete solution. In the case of long-term non-contact with users, the intelligent verification service is used to verify the user's connectability. If the user is no longer connectable, the lost connection repair service is used to find the current contact information and then contact the user. When there is a need to activate existing users or attract new users, the intelligent contact service is used to accurately screen customer groups and reach out in a diversified manner. Each capability can provide standardized services in the form of interfaces, and the lost connection repair part also supports SaaS services to ensure that the overall solution can serve customers efficiently.

[0050] In the above detailed description, reference is made to the accompanying drawings, which form a part of this document. In the accompanying drawings, similar symbols typically identify similar components unless the context indicates otherwise. The illustrated embodiments described in the detailed description, drawings, and claims are not meant to be limiting. Other embodiments may be used, and other changes may be made, without departing from the spirit or scope of the subject matter presented herein.

[0051] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A smart connection method based on big data technology, characterized in that: The authorized and compliant user identification requirements are encrypted and transmitted to the platform. The platform establishes a user model based on the user identification through a big data model algorithm, and completes the intelligent connection and contact step based on the user model combined with the contact capability. The intelligent connection and contact step further includes an intelligent connection verification step, a lost connection repair step, and an intelligent arrival step; In the Zhilian verification step, the customer transmits the irreversibly encrypted user identification information through the interface, matches the existing Zhilian verification model according to the transmitted user identification information, and returns the verification result to the user, and determines whether the user is connectable according to the verification result. If the user is not connectable, repair is recommended. If the user is connectable, the degree of connectability is further determined to be high, medium or low; The lost contact repair step constructs a repair model based on multi-dimensional data. The repair model determines that the repaired contact information is a phone number that has been online recently, removes recently inactive contact information, ensures the reach rate, and at the same time determines the user's complaint tendency, controls the complaint rate after the reach, outputs the user's contact information, and in the reach stage after the repair, uses China Unicom's privacy number capability to protect user privacy and does not return the real number; The intelligent reaching step builds user screening models for different industries based on the user's basic information, behavioral data, and consumption data in combination with different scenarios, performs secondary processing and in-depth analysis on the data, screens potential high-willing users, completes user portraits, and enables diversified reach.

2. The intelligent connection method based on big data technology as claimed in claim 1, characterized in that: The intelligent connection verification model verifies whether the mobile phone number belongs to the user based on different dimensional data such as the user's network access date, online status, activity in the past three months, online time and secondary card verification, and uses a logistic regression algorithm to construct the user's connection score and repair score. The connection score is used to score whether the user is reachable, and the connection is divided according to the connection score and the result is returned; whether the connection is possible is returned according to the connection score. If the connection is not possible, whether repair is recommended is returned according to whether there is an online contact method under the user's name; the repair score returns whether repair is recommended.

3. The intelligent connection method based on big data technology as claimed in claim 2, characterized in that: The Zhilian validation model further includes data selection, data processing, IV calculation and feature selection, logistic regression algorithm and parameter tuning, validation set evaluation and scoring model output; The data selection selects user data of a single month, and based on user samples, classifies 34 feature fields and 1 label field, and divides the data into 8:

2. 80% of the data will be used for training the connectable model, and the remaining 20% ​​of the data will be used for performance testing of the connectable model and reference setting of the scoring threshold, to examine the generalization ability of the connectable model and the rationality of the connectable parameter setting; The data processing further includes missing value filling, feature binning and WOE binning. When the field value type of the missing value is discrete, a random forest classifier is used to fill it. When the field value type of the missing value is continuous, a random forest regressor is used to fill it. The chi-square binning method is used to perform feature binning on continuous variables to achieve data discretization. For each bin or each value of the discrete variable, the woe value is calculated by the following formula (1) to perform WOE binning: （1） Bad total Indicates the total number of bad samples, Bad X=Xi represents bad samples and X is the number of the i-th group, Good X=Xi Good represents good samples and X is the number of the i-th group; total Indicates the total number of good samples; The IV calculation and feature selection are performed to screen variables through the linear correlation between features and the IV value of features: the stronger the correlation between features, the stronger the feature collinearity will be, and the greater the impact on label prediction will be. Before modeling, try to remove redundant features, that is, features that have strong correlation with other features but little contribution to the label; The larger the IV value, the greater the contribution of the feature to the label amount. The input parameters and features are ultimately determined based on the IV value and feature selection. The logistic regression algorithm and parameter tuning, using feature data to train the logistic regression algorithm, setting a logistic regression algorithm parameter search grid, using the area under the curve AUC value as a reference for algorithm performance, using a random search algorithm to optimize the algorithm parameters to obtain the optimal parameter set; the AUC value ranges from 0 to 1, and the larger the value, the better the model classification performance; The validation set evaluation uses a validation set accounting for 20% to evaluate the model effect, uses the AUC value to calculate the AUC values ​​of the training samples and the test samples respectively, and then evaluates whether the model can be put into production based on the AUC value; The scoring model output converts the model into a score output, and the connectability is divided into high and low levels according to the score. The final result is the connectability level of the input user.

4. The intelligent connection method based on big data technology as claimed in claim 3, characterized in that: The WOE binning further includes two bins for the number of active times: [5000, 10000) and [10000, 30000). The woe of the former bin is 2, and the latter is -1, indicating that users active in [10000, 30000) are unlikely to be unreachable, while users in [5000, 10000) are more likely to be unreachable.

5. The intelligent connection method based on big data technology as claimed in claim 4, characterized in that: The evaluation of whether the model can be put into production based on the AUC value further includes: AUC value > 0.63: the model is distinguishable on whether it can be connected, but the distinguishability is not sufficient to meet the verification scenario requirements of intelligent connection verification and cannot be put into production; AUC value > 0.73: the model is excellent and can be put into production; AUC value: 0.8029, discrimination evaluation index KS: 0.4500, accuracy: 0.7954 meet the production requirements.

6. The intelligent connection method based on big data technology as claimed in claim 2, characterized in that: The smart arrival step further comprises: Establish a data screening rule model, which deletes invalid users and non-mobile network users, and only retains the mobile network and online users that are currently in normal use; set a network access time threshold to exclude users who have been on the network for less than six months from the promotion candidate list, aiming to focus promotion resources more on stable users who have established certain usage habits; introduce a "three-no" user identification algorithm to identify users who have no recent call, SMS or data traffic usage records, and remove them from the promotion set; A user screening model is established to screen out users who meet the requirements. The user screening model first performs fine division based on the user's age information, and only retains the user group between the ages of 22 and 45. Then, the user screening model screens out users who have recently visited the relevant website more than 1 time and less than 8 times, and sorts them from the recent to the longest visit time and the low to high visit times to obtain a user result set. Obtain a user screening model for operator product cloud phones, and screen users who have demand and willingness for cloud phone operator product projects based on user interaction information and the user's own attribute characteristics; perform personalized processing based on the unique characteristics of the target users of the product cloud phone.

7. The intelligent connection method based on big data technology as claimed in claim 6, characterized in that: The user screening model for cloud phones of operator products further includes: for users who frequently participate in certain activities, deeply mining the user behavior data in the past month, comprehensively considering the click frequency and stay time of users on certain applications, and accurately screening out high-frequency user groups based on intelligent sorting algorithms to form a target user set; for the target group of college students, setting an age threshold, limiting the user age to under 30 years old as a preliminary screening condition, using geographic location information, combined with electronic fence technology, judging whether the user is located in the campus area as the key basis for identifying students on campus, integrating traffic usage and learning application usage habits, and finally outputting highly accurate college student target results through comprehensive evaluation.

8. An intelligent connection and contact system based on big data technology, characterized in that: The platform accepts the encrypted transmission of authorized and compliant user identification requirements. The platform establishes a user model based on the user identification through a big data model algorithm, and completes the intelligent connection and contact module based on the user model combined with the contact capability. The intelligent connection and contact module further includes an intelligent connection verification submodule, a lost connection repair submodule and an intelligent arrival submodule; The Zhilian verification submodule matches the existing Zhilian verification model according to the irreversibly encrypted user identification information transmitted by the customer, and returns the verification result to the user, and determines whether the user is connectable according to the acceptance result. If the user is not connectable, it recommends repair. If the user is connectable, it further determines whether the degree of connectability is high, medium or low. The lost contact repair submodule builds a repair model based on multi-dimensional data. The repair model determines that the contact information to be repaired is the phone number currently on the network, and removes the recently inactive contact information to ensure the reach rate. At the same time, it determines the user's complaint tendency and controls the complaint rate after the reach. The repair model outputs the user's contact information; in the reach stage after the repair, the China Unicom privacy number capability is used to protect the user's privacy and the real number is not returned; The intelligent reach submodule builds user screening models for different industries based on the user's basic information, behavioral data and consumption data, combined with different scenarios, and performs secondary processing and in-depth analysis on the data to screen potential high-willing users, complete user portraits, and enable diversified reach.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, the steps of the intelligent connection method based on big data technology described in any one of claims 1 to 7 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by the processor, the steps of the intelligent connection method based on big data technology described in any one of claims 1 to 7 are implemented.

Images