Property fee collection system based on large model

Abstract

The invention discloses a property fee collection system based on a large model, and relates to the technical field of large model and property management, and the system comprises a data collection and integration module, a large model training and optimization module, a collection strategy generation module, a collection execution and feedback module, and a monitoring and evaluation module. According to the method, multi-source heterogeneous data can be rapidly analyzed through a large model technology, arrearage reasons and repayment willingness are accurately predicted, personalized collection strategies are dynamically generated, the collection process is adjusted and optimized in real time, the collection efficiency and success rate are remarkably improved, meanwhile, it is ensured that collection behaviors are legal and compliant, and the relationship between property management and owners is improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of large models and property management, and specifically relates to a property fee collection system based on a large model. BACKGROUND

[0002] In the field of property management, property fee and related fee collection is an important and challenging task. Traditional collection methods mainly include offline and online forms. Offline collection usually relies on property management personnel to collect on-site or post collection notices at the door of the owner. This method not only consumes a lot of manpower and time cost, but also has low efficiency, and is ineffective for owners who are away for a long time or deliberately avoid it. Frequent on-site collection may also cause the owner's resentment, affecting the relationship between the property and the owner. Online collection is mostly limited to sending collection information or point-to-point communication collection, but the collection information is easy to be ignored by the owner, and point-to-point communication may not effectively convey the collection content due to the owner's shielding of the message or not checking in time. In addition, the existing collection methods generally lack in-depth analysis of the specific situation of the owner, making it difficult to achieve personalized collection. Different owners have different reasons for not paying, such as financial difficulties, dissatisfaction with services, disputes, forgetfulness, or lack of understanding of property service content, etc., and traditional collection methods cannot take effective measures to address these differences, resulting in low collection success rate.

[0003] With the development of artificial intelligence technology, especially the breakthrough of large model technology in natural language processing and data analysis, a new solution has been provided for property collection work. Large models have powerful data processing capabilities and natural language understanding capabilities, which can quickly analyze massive data and accurately locate the reasons for arrears and the willingness of the owner to repay, thereby tailoring personalized collection strategies for each owner. However, the current property collection field has not fully utilized the advantages of large model technology, and still has problems such as incomplete data collection, insufficient analysis depth, and single collection strategy. Therefore, how to deeply integrate large model technology with the property collection scenario and build an automated, differentiated, and compliant intelligent collection system has become a technical problem to be solved. The present application aims to provide a property fee collection system and method based on a large model by integrating large language models, voice interaction technology, and compliance knowledge base, to solve the problems of low efficiency and poor targeting of existing collection methods, improve the efficiency and success rate of collection work, and improve the relationship between the property and the owner. SUMMARY

[0004] The present application aims to provide a property fee collection system based on a large model, mainly to solve the problem of low efficiency and poor targeting of existing collection methods.

[0005] To achieve the above-mentioned purpose, the technical solution adopted by the present application is as follows:

[0006] The utility fee collection system based on a large model comprises a data collection and integration module, a large model training and optimization module, a collection strategy generation module, a collection execution and feedback module, and a monitoring and evaluation module; wherein the data collection and integration module is used to acquire and clean data from internal and external data sources, the large model training and optimization module trains a large model through historical data processed by the data collection and integration module to predict the reasons for arrears and the repayment willingness, the collection strategy generation module dynamically generates individualized collection strategies according to the output of the large model training and optimization module, the collection execution and feedback module is responsible for executing the generated individualized collection strategies and recording feedback information in real time, and the monitoring and evaluation module is used to quantify the collection effect and continuously optimize the system performance.

[0007] Further, in the present application, the data collection and integration module comprises an internal data collection unit, an external data collection unit, and a data cleaning and conversion unit; the internal data collection unit extracts the owner basic information, payment records, communication records, and collection results from the external property management system, financial charging system, customer service system, and collection system; the external data collection unit calls macroeconomic data and real estate market data through an API interface, and uses a web crawler to capture the surrounding environment dynamics and policies and regulations of the community.

[0008] Further, in the present application, the data cleaning and conversion unit removes duplicate records, fills in missing values with mean, median or mode, corrects or deletes abnormal values through statistical analysis and business rule judgment, processes text data through word segmentation and TF-IDF vectorization, and completes standardization coding by a text processing unit, normalizes numerical data to the 0-1 interval by a numerical processing unit, processes classification data by a classification processing unit using one-hot encoding or label encoding, and associates internal and external data by the owner ID or house number through a data association module.

[0009] Further, in the present application, the large model training and optimization module comprises a data annotation and division unit, a model architecture design unit, and a model training and updating unit; the data annotation and division unit annotates historical data and divides the reasons for arrears into four categories of service quality dissatisfaction, forgotten payment, economic difficulties, and service dispute;

[0010] The model architecture design unit designs a deep learning model comprising a text layer, a numerical layer, and an output layer, wherein the text layer processes the semantic features of the communication records through a bidirectional LSTM, the numerical layer processes the change trend of the payment data through a time series model, and the output layer combines classification tasks and regression tasks to predict the reasons for arrears and the repayment willingness, respectively;

[0011] The model training and update unit scores repayment willingness on a 5-point scale and classifies payment results into three states: success, failure, and installment. These are then divided into training set, validation set, and test set in a 7:2:1 ratio to train and update the model.

[0012] Furthermore, in this invention, the collection strategy generation module includes a debt reason analysis unit, a repayment willingness assessment unit, and a strategy formulation unit; the debt reason analysis unit identifies the dissatisfaction and demands expressed by homeowners through sentiment analysis and keyword matching, and infers the reasons for debt by combining payment records and house usage; the repayment willingness assessment unit extracts features such as historical payment behavior, communication attitude, and economic status, and calculates the repayment willingness score by weighted average.

[0013] Furthermore, in this invention, the strategy formulation unit obtains the total amount of property fees Pz currently payable by the homeowner and compares it with a set standard value Pc. If Pz > Pc, the collection execution and feedback module is triggered to send a reminder message to the homeowner to pay the property fees. If Pz ≤ Pc, a payment voucher corresponding to the homeowner is generated and stored in the collection execution and feedback module, and a reminder message that a payment voucher has been generated is sent to the homeowner. When generating the payment voucher corresponding to the homeowner, the specific value of the total amount of property fees Pz currently payable by the homeowner being less than Pc is also determined. When the specific value reaches a set threshold, a specific identifier is attached to the generated payment voucher. This specific identifier is used to adjust the priority and intensity of subsequent collection strategies.

[0014] The strategy formulation unit generates personalized collection strategies based on different reasons for overdue payments. For homeowners dissatisfied with service quality, it provides solutions and complimentary services as compensation. For homeowners facing financial difficulties, it requests the formulation of installment repayment plans and adjusts the amount and term. For homeowners who forget to pay, it sends gentle reminders via SMS and the app. For homeowners with service disputes, it cites laws and regulations to clarify the boundaries of responsibility and provides the results of similar case handling.

[0015] Furthermore, in this invention, the collection execution and feedback module includes a collection execution unit and a feedback adjustment unit. The collection execution unit first calculates the priority of the collection strategy. Based on the statistical values ​​of the community's basic parameters and homeowner behavior data collected by the internal data acquisition unit, it extracts the matching calculated values ​​from the corresponding lookup table and combines them with the level value corresponding to the specific identifier of the payment voucher, and performs a comprehensive calculation according to the following formula:

[0016] S=α×G+β1×(V pd +V ps +V cn +V cr +V cf )+β2×(V fp +V fpavg +V fa +Vfaavg +V fe +V feavg )

[0017] Wherein, S represents the priority of the collection strategy, alpha represents the adjustment factor, G represents the grade value corresponding to the specific identifier of the payment certificate, V pd represents the operation value of the property fee unit price, V ps represents the operation value of the property fee collection rate, V cn represents the operation value of the cell construction period, V cr represents the operation value of the cell area, V cf represents the operation value of the cell supporting facility condition, V fp represents the operation value of the payment frequency of the current period, V fpavg represents the average value of the operation value of the payment frequency of the past period, V fa represents the operation value of the total payment amount of the current period, V faavg represents the average value of the operation value of the total payment amount of the past period, V fe represents the operation value of the payment evaluation score, V feavg represents the average value of the operation value of the payment evaluation score of the past period, beta1 and beta2 are weight coefficients, and beta1+beta2=1;

[0018] The collection execution unit selects the best collection mode according to the calculated priority value and the repayment willingness score, including the short message and WeChat collection mode, the telephone collection mode, the on-site collection mode, the past collection notice mode, the legal notice collection mode, the lawyer notice collection mode, the court lawsuit collection mode and the third-party collection mode.

[0019] Further, in the present application, the feedback adjustment unit real-time inputs the owner feedback information, including opinions, newly raised questions and promised payment time, and transmits to the large model to reevaluate the arrears reason and repayment willingness, and dynamically adjusts the collection strategy according to the feedback information.

[0020] Further, in the present application, the monitoring and evaluation module includes a collection progress monitoring unit and an effect evaluation unit, the collection progress monitoring unit establishes a real-time updated collection progress database, and displays the collection progress in the form of column chart, line chart and progress bar, the effect evaluation unit regularly calculates the collection success rate and the recovery rate, the collection success rate is the number of successfully collected owners divided by the total number of arrears owners multiplied by 100%, and the recovery rate is the actual recovered arrears amount divided by the total arrears amount multiplied by 100%.

[0021] Compared with the prior art, the present application has the following beneficial effects:

[0022] (1) The present application can accurately locate the reasons for arrears and repayment willingness through the rapid analysis of a large number of owner data by a large model, thereby tailoring the collection strategy for each owner, avoiding the blindness of traditional collection methods, greatly improving the efficiency of collection work, and saving manpower and time cost.

[0023] (2) The personalized collection strategy of the present application can better meet the needs of different owners, take appropriate measures according to the specific circumstances of the owners, and more easily persuade the owners to pay. At the same time, the collection strategy is adjusted in time according to the feedback information, further improving the success rate of collection and increasing the income of the property.

[0024] (3) The collection system of the present application adopts a more personalized and targeted collection method, avoiding the disturbance and aversion to the owners caused by excessive or improper collection, which helps to improve the relationship between the property and the owners and enhance the satisfaction and trust of the owners to the property, laying a good foundation for subsequent property management work.

[0025] (4) The monitoring and evaluation of the present application can quantitatively analyze the collection effect, providing data support for the optimization of the large model and the improvement of the collection strategy. With the continuous accumulation of data and in-depth analysis, the system can continuously adapt to new situations and continuously improve the quality and effect of the collection work. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is the overall structure principle diagram of the present application.

[0027] Figure 2 is the working schematic diagram of the data collection and integration module in the present application.

[0028] Figure 3 is the working schematic diagram of the large model training and optimization module in the present application. DETAILED DESCRIPTION

[0029] The present application will be further described below in combination with the accompanying drawings and examples. The modes of the present application include but are not limited to the following examples.

[0030] EXAMPLE

[0031] The present application discloses a property fee collection system based on a large model, which is combined with the accompanying drawings Figure 1 to Figure 3 , the specific implementation process of the present application is described in detail. The following content will be expanded according to five modules of data collection and integration, large model training and optimization, collection strategy generation, collection execution and feedback, and monitoring and evaluation, and each module and key step will be described one by one.

[0032] As Figure 2As shown, the data collection and integration module obtains information from multiple source heterogeneous data through internal and external data collection units. The internal data collection unit extracts owner basic information, payment records, communication records, and collection results from external property management systems, financial charging systems, customer service systems, and collection systems. These data are stored in a structured form in the database for subsequent processing. The external data collection unit calls macroeconomic data and real estate market data through API interface, and uses web crawlers to capture community environment dynamics and policies and regulations. For example, the external data collection unit can obtain the latest property fee payment related policies from government websites, or capture community activities or environmental changes related to the community from news platforms.

[0033] The data cleaning and conversion unit pre-processes the collected data, including de-duplication, missing value filling, outlier correction, and standardization coding. For duplicate records, a hash algorithm is used to identify and remove redundant data; for missing values, the mean, median, or mode is selected according to the field type to fill; for outliers, they are corrected or deleted after statistical analysis and business rule judgment. Text data is processed by word segmentation and TF-IDF vectorization to complete standardization coding, numerical data is normalized to the 0-1 interval, and categorical data is encoded using one-hot encoding or label encoding. In addition, the data correlation module correlates internal and external data through owner ID or house number to ensure data consistency and integrity.

[0034] The large model training and optimization module includes data labeling and division unit, model architecture design unit, and model training and update unit. As shown, Figure 3 The data labeling and division unit labels historical data, classifying overdue payment reasons into four categories: service quality dissatisfaction, forgetfulness, economic difficulties, and service disputes. Repayment willingness is rated on a 5-point scale, and payment results are classified into three states: success, failure, and installment. All labeled data are divided into training set, validation set and test set in the ratio of 7:2:1.

[0035] According to the repayment willingness score, overdue owners are divided into high repayment willingness (corresponding to the score interval of 4-5), medium repayment willingness (corresponding to the score interval of 2-4), and low repayment willingness (corresponding to the score interval of 1-2).

[0036] High repayment willingness

[0037] For owners with high repayment willingness, encouragement and support are the main focus. In the collection strategy, express recognition and appreciation of the owner's positive attitude, provide convenient payment channels and methods such as online payment, self-payment, etc., and give certain payment convenience such as extension of payment period, provision of payment service fee reduction, etc. At the same time, establish a good communication mechanism, answer the owner's questions in a timely manner, provide necessary help and support, and ensure that the owner can complete the payment smoothly.

[0038] Middle repayment willingness

[0039] For homeowners with medium repayment willingness, strengthen communication and guidance during the collection process. Through phone calls, face-to-face communication, and other means, deeply understand the homeowner's thoughts and difficulties, and solve the homeowner's problems and concerns. Increase some incentives in the collection strategy, such as giving certain points rewards, service discounts, etc., to improve the homeowner's payment enthusiasm. At the same time, clearly inform the homeowner of the consequences and impact of arrears, such as damaged credit record, generated late fees, etc., to prompt the homeowner to pay as soon as possible.

[0040] Low repayment willingness

[0041] For homeowners with low repayment willingness, take a more tough and formal collection approach. First, through phone calls, written notices, etc., inform the homeowner of the facts and consequences of arrears, and clearly require the homeowner to provide a clear repayment plan and commitment within the specified time. If the homeowner still does not cooperate, consider sending a lawyer's letter or taking legal action to protect the legal rights and interests of the property. The property service time, content, and collection time, method, and content can be automatically generated in the intelligent property management system to form evidence materials and can be exported for printing, ensuring the legality and standardization of the collection behavior. At the same time, do not give up communication with the homeowner, try to understand the deep reasons for not paying, and find possible solutions to avoid the intensification of conflicts.

[0042] For example: In the repayment willingness assessment unit, the system extracts the homeowner's characteristics (historical payment behavior, communication attitude, economic status), and calculates the repayment willingness score (5-point system) through weighted average. The weights are determined based on historical data statistics. The following is an example of a homeowner's data:

[0043] (1) Historical payment behavior score: 4.0 (based on timely payment ratio)

[0044] (2) Communication attitude score: 3.5 (based on sentiment analysis of customer service communication records)

[0045] (3) Economic status score: 2.5 (based on external economic data evaluation)

[0046] (4) Weight distribution (according to historical statistics): historical payment behavior weight = 0.5, communication attitude weight = 0.3, economic status weight = 0.2 (the total weight is 1).

[0047] Calculation process: Repayment willingness score = (historical payment behavior score × weight) + (communication attitude score × weight) + (economic status score × weight) = (4.0 × 0.5) + (3.5 × 0.3) + (2.5 × 0.2) = 2.0 + 1.05 + 0.5 = 3.55

[0048] Result: The repayment willingness score of this property owner is 3.55 (out of 5). The system classifies this property owner as a medium repayment willingness property owner and uses this score to generate a personalized collection strategy (e.g., a lower score may trigger more aggressive collection methods).

[0049] The model architecture design unit selects a suitable large model architecture based on the characteristics of the property collection data and the analysis requirements. For example, the Transformer architecture is chosen because it performs well in natural language processing and data analysis tasks, and can effectively handle sequential data and capture complex relationships between data. Based on the selected architecture, the specific structure of the model is designed, and the parameters of the input layer, hidden layer, and output layer are determined. For example, the number of input layer neurons corresponds to the dimension of the data features, and the number of layers and neurons in the hidden layer is adjusted according to the complexity of the task and the amount of data to ensure that the model has sufficient expression ability.

[0050] This embodiment designs a deep learning model that includes a text layer, a numerical layer, and an output layer. The text layer processes the semantic features of the communication records using a bidirectional LSTM, the numerical layer processes the trend of payment data using a time series model, and the output layer combines classification and regression tasks to predict the reasons for arrears and repayment willingness, respectively. The model training and updating unit uses a multi-task loss function combining cross-entropy and mean square error for optimization, with an initial learning rate of 1e-4, which is reduced by 50% when the validation set loss value no longer decreases. The training period is set to 50 rounds, and the optimal parameter model is saved. For real-time new data, the model uses an incremental training strategy to optimize the new sensitive layer while preserving historical parameters, thereby improving the model's generalization ability.

[0051] The collection strategy generation module includes an arrears reason analysis unit, a repayment willingness evaluation unit, and a strategy formulation unit. The arrears reason analysis unit identifies the owner's expression of dissatisfaction and appeal information through sentiment analysis and keyword matching, and infers the reasons for arrears based on payment records and housing usage. The large model extracts features from the owner's historical payment behavior, such as the timeliness of payment, the frequency of arrears, the duration of each arrears, etc. At the same time, it analyzes the language expression and attitude of the owner when communicating with the property. For example, if the owner shows a sincere attitude and actively proposes a solution, such as requesting a delay in payment, a discount appeal, or developing a repayment plan, the large model will give a higher repayment willingness score; if the owner ignores the collection information or has a bad attitude, the repayment willingness score will be lower.

[0052] The extracted features are quantitatively processed and assigned different weights through the establishment of a mathematical model, and the repayment willingness evaluation result of the property owner is calculated comprehensively. For example, the weight of payment timeliness can be set to 0.5, the weight of communication attitude to 0.3, and the weight of arrears frequency to 0.2, and the repayment willingness score is calculated by weighted average method.

[0053] The policy making unit obtains the total amount of property fees Pz that the owner should currently pay, and compares it with the set standard value Pc. If Pz>Pc, the collection execution and feedback module is triggered to send a message to the owner reminding him to pay the property fees; if Pz≤Pc, a payment voucher corresponding to the owner is generated and stored in the collection execution and feedback module, and a reminder message is sent to the owner that a payment voucher has been generated. The generated payment voucher is attached with a specific identifier, which is used to adjust the priority and intensity of the subsequent collection strategy. The policy making unit generates individualized collection strategies according to different reasons for arrears: for owners who are not satisfied with the quality of service, provide solutions to problems and offer service compensation; for owners who are economically disadvantaged, develop a installment repayment plan and adjust the amount and period; for owners who forget to pay, send a gentle reminder through SMS or APP push; for owners who dispute the service, cite legal regulations to clearly define the responsibility limits and provide similar case handling results.

[0054] The collection execution and feedback module includes a collection execution unit and a feedback adjustment unit. The collection execution unit first calculates the priority of the collection strategy, the formula is as follows: S=α×G+β1×(V pd +V ps +V cn +V cr +V cf )+β2×(V fp +V fpavg +V fa +V faavg +V fe +V feavg ); where S represents the priority of the collection strategy, α represents the adjustment factor, G represents the grade value corresponding to the specific identifier of the payment voucher, V pd represents the calculated value of the property fee unit price, V ps represents the calculated value of the property fee collection rate, V cn represents the calculated value of the age of the community, V cr represents the calculated value of the region to which the community belongs, V cf represents the calculated value of the community supporting facilities, V fp represents the calculated value of the current payment frequency, V fpavg represents the average value of the calculated value of the past payment frequency, V fa represents the calculated value of the current total payment amount, V faavg represents the average value of the calculated value of the past total payment amount, V fe represents the calculated value of the payment evaluation score, V feavg represents the average value of the calculated value of the past payment evaluation score, β1 and β2 are weight coefficients, and β1+β2=1.

[0055] The following is an example of data for a community:

[0056] a = 0.1 (adjustment factor);

[0057] G = 0.8 (specific identification level value of payment certificate, high level);

[0058] b1 = 0.6, b2 = 0.4 (weighting factor);

[0059] First group of variables (cell characteristics):

[0060] V pd = 0.7 (property fee unit price calculation value);

[0061] V ps = 0.6 (property fee collection rate calculation value);

[0062] V cn = 0.4 (cell construction age calculation value);

[0063] V cr = 0.5 (cell area calculation value);

[0064] V cf = 0.8 (cell supporting facilities calculation value).

[0065] Second group of variables (property owner payment behavior):

[0066] V fp = 0.9 (current payment frequency calculation value);

[0067] V fpavg = 0.7 (average of past payment frequency calculation values);

[0068] V fa = 0.8 (current total payment amount calculation value);

[0069] V faavg = 0.75 (average of past total payment amount calculation values);

[0070] V fe = 0.85 (current payment evaluation score calculation value);

[0071] V feavg = 0.8 (average of past payment evaluation score calculation values).

[0072] Calculation process:

[0073] 1. First group and = V pd + V ps + V cn + V cr + V cf = 0.7 + 0.6 + 0.4 + 0.5 + 0.8 = 3.0

[0074] 2. Second group sum = V fp + V fpavg + V fa + V faavg + V fe + V feavg = 0.9 + 0.7 + 0.8 + 0.75

[0075] + 0.85 + 0.8 = 4.8

[0076] 3. S = a x G + b1 x {first group sum} + b2 x {second group sum} 0.1 x 0.8 + 0.6 x 3.0 + 0.4 x 4.8

[0077] = 0.08 + 1.8 + 1.92 = 3.8

[0078] Result: priority S = 3.8. The system selects the collection method according to this value (for example, higher S may prioritize phone collection or on-site collection).

[0079] The collection execution unit selects the optimal collection method according to the calculated priority value, including SMS and WeChat collection methods, phone collection methods, on-site collection methods, posting collection notices, legal notices, lawyer letters, court proceedings, and third-party collection methods. The feedback adjustment unit real-time enters the owner feedback information, including opinions, new questions, and commitment payment time, and transmits to the large model to re-evaluate the reasons for arrears and repayment willingness, and dynamically adjusts the collection strategy according to the feedback information. For example, if the owner feedbacks that he needs to delay payment due to economic difficulties, the system automatically adjusts the collection strategy to a installment repayment plan.

[0080] The monitoring and evaluation module includes a collection progress monitoring unit and an effect evaluation unit. The collection progress monitoring unit establishes a real-time updated collection progress database, and displays the collection progress in the form of column chart, line chart and progress bar. The effect evaluation unit regularly calculates the collection success rate and recovery rate. The collection success rate is the number of successfully collected owners divided by the total number of arrears owners multiplied by 100%, and the recovery rate is the actual recovered arrears amount divided by the total arrears amount multiplied by 100%. In addition, the effect evaluation unit also generates detailed collection reports, including the execution effect of the collection strategy, the distribution of owner feedback and system optimization suggestions, providing data support for subsequent improvement.

[0081] The effect evaluation unit regularly calculates the collection success rate and recovery rate, the formula is as follows:

[0082] Collection success rate = (number of successfully collected owners / total number of arrears owners) x 100%;

[0083] Recovery rate = (actual recovered arrears amount / total arrears amount) x 100%;

[0084] Example data (assuming one collection cycle has ended):

[0085] Number of delinquent property owners: 100;

[0086] Number of successfully collected property owners: 75 (defined as having paid off the debt or reached a repayment agreement);

[0087] Total amount of outstanding debt: 50,000 yuan;

[0088] Actual amount recovered: 40,000 yuan (including partial payments and installment payments);

[0089] Calculation process:

[0090] 1. Collection success rate = (75 / 100) x 100% = 75%

[0091] 2. Recovery rate = (40,000 / 50,000) x 100% = 80%

[0092] Result: The collection success rate is 75%, and the recovery rate is 80%. The system generates a report for optimizing subsequent strategies (for example, if the recovery rate is low, the collection intensity or strategy type may be adjusted).

[0093] In summary, the present application constructs an intelligent collection system that is automated, differentiated, and compliant by deeply integrating large model technology, significantly improving the efficiency and success rate of property collection work. The system can be flexibly adjusted according to different scenarios in actual application, such as achieving efficient collection through batch processing in large-scale communities and improving property owner satisfaction through customized services in high-end residential areas. Through the above specific implementation, the present application realizes full-process closed-loop management from data collection to strategy generation and execution feedback, solves the technical problems of low efficiency and poor targeting of traditional collection methods, and has significant technical progress and application value.

[0094] The above embodiment is only one of the preferred embodiments of the present application and should not be used to limit the protection scope of the present application. Any modification or polishing made within the main design idea and spirit of the present application without substantial meaning, which still solves the technical problems consistent with the present application, should be included in the protection scope of the present application.

Claims

1. A large model-based property fee collection system, characterized by, The system comprises a data collection and integration module, a large model training and optimization module, a collection strategy generation module, a collection execution and feedback module, and a monitoring and evaluation module.

2. The large model-based property fee collection system according to claim 1, wherein, The data collection and integration module comprises an internal data collection unit, an external data collection unit, and a data cleaning and conversion unit.

3. The large model-based property fee collection system of claim 2, wherein, The data cleaning and conversion unit removes duplicate records, fills in missing values using mean, median or mode, corrects or deletes abnormal values through statistical analysis and business rules, processes text data through word segmentation and TF-IDF vectorization, and completes standardization coding by a text processing unit, normalizes numerical data to the 0-1 interval by a numerical processing unit, processes categorical data using one-hot encoding or label encoding by a categorical processing unit, and associates internal and external data by a data association module.

4. The large model-based property fee collection system according to claim 2, wherein, The large model training and optimization module comprises a data annotation and division unit, a model architecture design unit, and a model training and updating unit. The data annotation and division unit annotates historical data and classifies the reasons for arrears into four categories: service quality dissatisfaction, forgetfulness, economic difficulties, and service disputes. The model architecture design unit designs a deep learning model containing a text layer, a numerical layer, and an output layer. The model training and updating unit scores the repayment intention on a 5-point scale, classifies the payment results into success, failure, and installment, and divides them into training set, validation set, and test set in a 7:2:1 ratio to train and update the model.

5. The large model-based property fee collection system according to claim 4, wherein, The collection strategy generation module comprises a reason analysis unit, a repayment intention evaluation unit, and a strategy formulation unit. The reason analysis unit identifies the owner's dissatisfaction and appeal information through sentiment analysis and keyword matching, and infers the reasons for arrears based on payment records and housing usage. The repayment intention evaluation unit extracts historical payment behavior, communication attitude, and economic status to calculate the repayment intention score.

6. The large model-based property fee collection system according to claim 5, wherein, The policy making unit obtains the total amount of property fees Pz currently payable by the owner and compares it with the set standard value Pc, if Pz>Pc, the collection execution and feedback module is triggered to send a message to the owner reminding him to pay the property fees, if Pz≤Pc, a payment voucher corresponding to the owner is generated and stored in the collection execution and feedback module, and a reminder message that the payment voucher has been generated is sent to the owner; wherein when the payment voucher corresponding to the owner is generated, it is also judged that the specific value of the total amount of property fees Pz currently payable by the owner is less than Pc, when the specific value reaches the set threshold, a specific identifier is attached to the generated payment voucher, which is used to adjust the priority and intensity of the subsequent collection strategy; The policy making unit generates individualized collection strategies according to different arrears reasons, provides problem solving solutions and service compensation for owners who are not satisfied with the service quality, requires the development of installment payment plan and the adjustment of amount and period for owners who are economically difficult, gently reminds the owners who forget to pay through SMS and APP push, and quotes laws and regulations to clarify the responsibility limit and provide similar case processing results for service dispute owners.

7. The large model-based property fee collection system of claim 6, wherein, The collection execution and feedback module includes a collection execution unit and a feedback adjustment unit, the collection execution unit first calculates the priority of the collection strategy, according to the statistical value of the community basic parameters and the owner behavior data collected by the internal data acquisition unit, extracts the matching operation value from the corresponding control table and combines the corresponding grade value of the specific identifier of the payment voucher, and calculates comprehensively according to the following formula: S = a x G + b1 x (V pd + V ps + V cn + V cr + V cf ) + b2 x (V fp + V fpavg + V fa + V faavg + V fe + V feavg ) wherein S represents a priority of a collection strategy, a represents an adjustment factor, G represents a grade value corresponding to a specific identifier of a payment certificate, V pd represents an operation value of a property fee unit price, V ps represents an operation value of a property fee collection rate, V cn represents an operation value of a cell construction age, V cr represents an operation value of a cell belonging area, V cf represents an operation value of a cell supporting facility condition, V fp represents an operation value of a payment frequency current period, V fpavg represents an operation value of a payment frequency past period average, V fa represents an operation value of a total payment amount current period, V faavg represents an operation value of a total payment amount past period average, V fe represents an operation value of a payment evaluation score, V feavg represents an operation value of a payment evaluation score past period average, β1 and β2 are weight coefficients, and β1 + β2 = 1; The collection execution unit selects the best collection method according to the calculated priority value and the repayment willingness score, including SMS and WeChat collection method, telephone collection method, on-site collection method, posting collection notice method, legal letter collection method, lawyer letter collection method, court prosecution collection method and third-party collection method.

8. The large model-based property fee collection system according to claim 7, wherein, The feedback adjustment unit inputs the owner feedback information in real time, including opinions, newly raised questions and promised payment time, transmits it to the large model to reevaluate the arrears reason and repayment willingness, and dynamically adjusts the collection strategy according to the feedback information.

9. The large model-based property fee collection system of claim 8, wherein, The monitoring and evaluation module includes a collection progress monitoring unit and an effect evaluation unit, the collection progress monitoring unit establishes a real-time updated collection progress database, and displays the collection progress in the form of column chart, line chart and progress bar, the effect evaluation unit regularly calculates the collection success rate and recovery rate, the collection success rate is the number of successfully collected owners divided by the total number of arrears owners multiplied by 100%, and the recovery rate is the actual recovered arrears amount divided by the total arrears amount multiplied by 100%.

Images