An AI social work-based visiting service implementation method and system

Abstract

The application relates to the technical field of artificial intelligence and social service, and discloses a visit service implementation method and system based on AI social work, wherein the method comprises the following steps: constructing a resident portrait vector and calculating a visit priority score through a graph attention network; adopting a Q-learning algorithm to optimize a visit path according to a spatial distance; obtaining conversation text and an emotional stress index through voice recognition and BERT semantic analysis; structuring and archiving visit data; and dynamically updating a resident portrait. Compared with the social work service mode in the prior art, which is mainly fixed scheduling visit and artificial experience judgment, especially under the conditions of diversified service population and implicit expression of risk emotions, the technical problem that precise visit service scheduling is difficult to realize can be solved, because the resident portrait priority is modeled through a graph neural network, and the path planning is driven by reinforcement learning, so that the social work task is efficiently scheduled, and the individualized coverage ability and intelligent decision-making ability of the social work service are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of artificial intelligence and social services, and particularly relates to a visit service implementation method and system based on AI social workers. BACKGROUND

[0002] At present, social worker visit services play an important role in urban and rural community governance, special population care and other scenarios, but their scheduling and management methods rely on manual experience, fixed scheduling and paper records, and have problems such as low efficiency, slow response, weak risk identification ability, and are difficult to adapt to diversified population service needs. For example, when facing high-risk groups such as the elderly living alone, emotional abnormalities, and psychological disorders, the existing technology cannot achieve accurate identification and priority response based on big data, resulting in serious lag in resource allocation and intervention timing. In addition, the information value of the conversation content is difficult to systematically extract, and the historical records are scattered in multiple terminals, lacking a unified archiving and portrait updating mechanism, which also makes it difficult to implement long-term service optimization. The existing technology cannot fully meet the complex needs of "multi-dimensional information fusion, implicit risk identification, task intelligent scheduling, and service closed-loop feedback" in social worker services. Therefore, there is an urgent need for an AI-driven visit service implementation method that can still achieve dynamic identification of high-risk groups, intelligent scheduling of task paths, structured archiving of interaction information, and continuous updating of resident portraits in large-scale distributed service scenarios, in order to improve the intelligent level and dynamic adaptability of social worker services. SUMMARY

[0003] In view of the above technical deficiencies, the purpose of the present application is to provide a visit service implementation method based on AI social workers, which aims to solve the technical problems that the existing social worker service methods are mostly fixed scheduling visits and manual experience judgments, especially under the conditions of diversified service population and implicit expression of risk emotions, it is difficult to achieve precise visit service scheduling.

[0004] To solve the above technical problems, the technical scheme adopted by the present application is as follows: the present application provides a visit service implementation method based on AI social workers,

[0005] The visit service implementation method based on AI social workers comprises:

[0006] Step S10: Collecting the basic attribute data, historical service record data and neighborhood interaction relationship data of community residents and establishing a resident portrait vector; pre-constructing and training a graph attention network model, inputting the resident portrait vector into the graph attention network model, and outputting the visit priority score of the resident;

[0007] Step S20: generating a high-priority resident set according to the visit priority score; obtaining the actual spatial distance, and calculating the optimized visit path by using the Q-learning reinforcement algorithm according to the high-priority resident set and the actual spatial distance;

[0008] Step S30: performing a visit process according to the optimized visit path; converting conversation content of the visit process into conversation text data in real time through voice recognition; calculating a resident emotional stress index based on the conversation text data in combination with a BERT semantic analysis method;

[0009] Step S40: structuring and archiving information using a graph database Neo4j according to the conversation text data and the resident emotional stress index, and constructing a visit data record unit;

[0010] Step S50: outputting a set of emotional risk residents according to the visit data record unit, and updating a resident portrait vector according to the visit data record unit.

[0011] Preferably, in step S10, the basic attribute data includes age data, gender data, residence status, marital status, and health status; the historical service record data includes service frequency, problem closure rate, and service satisfaction; the neighborhood interaction relationship data has a graph structure, and the neighborhood interaction relationship data includes spatial adjacency relationship, community event coordination relationship, community risk transmission factor, and adjacency node clustering coefficient of residents.

[0012] Preferably, in step S10, the structure of the graph attention network model specifically includes: an input layer for receiving the resident portrait vector; two hidden layers including a first layer hidden layer of 64 dimensions and a second layer hidden layer of 32 dimensions; a multi-head graph attention layer for attention weighted aggregation of each resident portrait vector and the resident portrait vectors of its neighbors; and a resident priority regression output layer adopting a Sigmoid activation function for outputting a visit priority score of the resident.

[0013] Preferably, in step S10, in the training process of the graph attention network model, a neighborhood interaction relationship graph structure consistency constraint is introduced to construct a loss function, and the loss function adopts a formula as follows:

[0014] ;

[0015] wherein, is the loss function; is the total number of residents in the training set in the training process; is a real priority label of the resident i; is a predicted priority label of the resident i predicted by the graph attention network model; is a neighborhood consistency regularization factor for controlling distribution smoothing guided by the graph structure; is a set of adjacency edges in the neighborhood interaction relationship data; is a real priority label of the resident j.

[0016] Preferably, in step S20, a high-priority resident set is generated according to the visit priority score; the actual spatial distance is obtained, and the Q-learning reinforcement algorithm is used to calculate the optimized visit path according to the high-priority resident set and the actual spatial distance, which specifically includes:

[0017] Step S201: preset a visit priority score threshold, and screen out a high-priority resident set with a visit priority score higher than the visit priority score threshold ;

[0018] Step S202: define the state space S of the Q-learning reinforcement algorithm, the state space S includes the resident path sequence of the completed visit, the current social worker position and the remaining visitable resident set; define the action space A of the Q-learning reinforcement algorithm, the action space A includes a set of next target residents selected from the current position; obtain the actual spatial distance, and design the reward function R of the Q-learning reinforcement algorithm according to the actual spatial distance and the high-priority resident set;

[0019] Step S203: according to the state space S, the action space A and the reward function R, the Q-learning reinforcement learning algorithm is used to perform iterative update, and the optimized visit path is output.

[0020] Preferably, in step S30, the visit process is performed according to the optimized visit path; the conversation content of the visit process is converted into conversation text data in real time through voice recognition; and the resident emotional stress index is calculated based on the conversation text data combined with the BERT semantic analysis method, which specifically includes:

[0021] Step S301: performing the visit process according to the optimized visit path, and using a mobile terminal to record the conversation content in the visit process in real time, and converting the voice stream data of the conversation content into conversation text data through an integrated voice recognition engine Wav2Vec 2.0 in real time;

[0022] Step S302: inputting the conversation text data into a pre-trained Chinese combined BERT semantic analysis method for semantic embedding coding to obtain a semantic vector of each round of conversation; setting a semantic risk keyword template set, and calculating a semantic similarity score of each round of conversation using a cosine similarity method;

[0023] Step S303: when the semantic similarity score exceeds a preset semantic similarity score threshold, a convolutional neural network is used to extract features from the semantic vector of the round of conversation to obtain emotion word feature recognition, emotional intensity feature, language repetition feature, speech speed change feature and word skipping frequency feature;

[0024] Step S304: Based on the emotion word feature recognition, the emotional intensity feature, the language repetition feature, the speech speed change feature and the word jump frequency feature, the resident emotional stress index is output by weighted linear combination.

[0025] Preferably, in step S50, the emotion risk resident set is output according to the visit data recording unit, and the resident portrait vector is updated according to the visit data recording unit.

[0026] Step S501: Extracting the sequence data of the emotional stress index of the resident and the sequence data of the conversation text data from the visit data recording unit;

[0027] Step S502: Risk feature extraction is performed on the sequence data of the emotional stress index of the resident and the sequence data of the conversation text data, and a risk feature vector is obtained, the risk feature vector including an emotional feature sub-vector, a text feature sub-vector and a service feedback feature sub-vector;

[0028] Step S503: Inputting the risk feature vector into a lightweight classifier, the lightweight classifier outputting a resident emotional risk level, and outputting residents with a resident emotional risk level greater than a preset resident emotional risk level threshold as an emotion risk resident set;

[0029] Step S504: Finally, updating the resident portrait vector according to the visit data recording unit.

[0030] The application also provides an AI social work-based visit service implementation system, which comprises:

[0031] The portrait construction module is used to collect the basic attribute data, the historical service record data and the neighborhood interaction relationship data of the community residents and establish a resident portrait vector; a graph attention network model is constructed and trained in advance, the resident portrait vector is input into the graph attention network model, and a visit priority score of the resident is output;

[0032] The path scheduling module is used to generate a high-priority resident set according to the visit priority score; an actual spatial distance is obtained, and a Q-learning reinforcement algorithm is used to calculate an optimized visit path according to the high-priority resident set and the actual spatial distance;

[0033] The visit execution and emotion analysis module is used to execute a visit process according to the optimized visit path; the conversation content of the visit process is converted into conversation text data in real time through voice recognition; and a resident emotional stress index is calculated based on the conversation text data and a BERT semantic analysis method;

[0034] The structured archiving module is used to structure and archive information by using a graph database Neo4j according to the conversation text data and the resident emotional stress index, and construct a visit data recording unit.

[0035] The risk identification and portrait updating module is configured to output a set of emotional risk residents according to the visit data recording unit and update a resident portrait vector according to the visit data recording unit.

[0036] The application further provides a computer program product comprising an AI-based social worker visit service implementation program, which, when executed by a processor, implements the AI-based social worker visit service implementation method.

[0037] The application has the advantages that, compared with the social worker service mode in the prior art, which mainly involves fixed schedule visits and artificial experience judgment, especially under the conditions of diversified service population and implicit expression of risk emotions, it is difficult to achieve precise visit service scheduling. The application realizes efficient scheduling of social worker tasks and improves the individualized coverage capability and intelligent decision-making capability of social worker services by modeling resident portrait priorities through a graph neural network and driving path planning through reinforcement learning. BRIEF DESCRIPTION OF DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0039] Figure 1 FIG. 1 is a flowchart of a first embodiment of the AI-based social worker visit service implementation method of the present application.

[0040] Figure 2 FIG. 2 is a schematic diagram of the equipment for the AI-based social worker visit service implementation method of the present application. DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments only constitute some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0042] Embodiment one: as shown in FIG. 1, a flowchart of a first embodiment of the AI-based social worker visit service implementation method of the present application is provided. Figure 1

[0043] In the first embodiment, the AI-based social worker visit service implementation method comprises:​

[0044] Step S10: Collecting basic attribute data, historical service record data and neighborhood interaction relationship data of community residents and establishing a resident portrait vector; a graph attention network model is constructed and trained in advance, the resident portrait vector is input into the graph attention network model, and a visit priority score of the resident is output;

[0045] It should be noted that in step S10, the basic attribute data includes age data, gender data, residence status, marital status and health status; the historical service record data includes service times, problem closure rate and service satisfaction; the data structure of the neighborhood interaction relationship data is a graph structure, and the neighborhood interaction relationship data includes the adjacency relationship of residents in space, the community event cooperation relationship, the community risk transmission factor and the adjacency node clustering coefficient. In step S10, in the training process of the graph attention network model, a neighborhood interaction relationship graph structure consistency constraint is introduced to construct a loss function, and the loss function adopts the formula:

[0046] ;

[0047] Wherein, is the loss function; is the total number of residents in the training set in the training process; is the true priority label of resident i; is the predicted priority label of resident i predicted by the graph attention network model; is a neighborhood consistency regularization factor for controlling the distribution smoothing guided by the graph structure; is a set of adjacent edges in the neighborhood interaction relationship data; is the true priority label of resident j.

[0048] It can be understood that the neighborhood interaction relationship graph structure consistency constraint regularization term introduced in the training of the above-mentioned graph attention network model makes the model not only learn the attribute characteristics of each resident node itself, but also perform joint modeling of information propagation and visit importance according to the high-risk commonness existing in the neighborhood relationship, thereby improving the identification ability of the model to potential implicit high-risk residents.

[0049] It should be understood that the traditional scoring method is mostly based on independent resident information and does not have the ability of spatial propagation and social group pattern recognition, which is easy to cause the missed judgment of residents "latent in the marginal area" or "not labeled but with concentrated neighborhood risk". The method automatically introduces social coordination signals through the edge connection relationship in the graph structure, so that the model can still produce a scoring result with consistent propagation under the condition that the resident label is incomplete or the service history is insufficient.

[0050] For example, in a certain community, residents A and B are upstairs and downstairs neighbors. Resident A has been identified as a mentally at-risk resident, and although there is no explicit service record for neighbor B, the model captures his high connection with A through the regularization term. The final model predicts his priority score as 0.87, while the traditional model only gets 0.42. Actual testing found that B was indeed a potential risk object due to emotional fluctuations in solitude, confirming the model's risk reasoning ability and identification accuracy driven by neighborhood relationships.

[0051] Step S20: generating a high-priority resident set according to the visit priority score; obtaining the actual spatial distance, and calculating the optimized visit path using the Q-learning reinforcement algorithm based on the high-priority resident set and the actual spatial distance;

[0052] It should be noted that in step S20, the high-priority resident set is generated according to the visit priority score; the actual spatial distance is obtained, and the optimized visit path is calculated using the Q-learning reinforcement algorithm based on the high-priority resident set and the actual spatial distance. The steps specifically include:

[0053] Step S201: presetting a visit priority score threshold, and screening out a high-priority resident set with a visit priority score higher than the visit priority score threshold ;

[0054] Step S202: defining the state space S of the Q-learning reinforcement algorithm, which includes the resident path sequence that has been completed, the current social worker location, and the remaining visitable resident set; defining the action space A of the Q-learning reinforcement algorithm, which includes the set of next target residents selected from the current location; obtaining the actual spatial distance, and designing the reward function R of the Q-learning reinforcement algorithm based on the actual spatial distance and the high-priority resident set;

[0055] Step S203: performing iterative updates using the Q-learning reinforcement learning algorithm based on the state space S, the action space A, and the reward function R, and outputting the optimized visit path.

[0056] It can be understood that by introducing the reinforcement learning mechanism, the social worker path planning is not only based on the resident priority score, but also dynamically adapts to the spatial distribution and resource cost, achieving the balance of "maximizing service value / minimizing distance". Compared with the static scheduling algorithm, Q-learning can optimize the strategy according to real-time feedback, adapt to personnel changes, resident lack of visit, and other uncertainties.

[0057] It should be understood that the traditional path optimization method (such as TSP, greedy algorithm) usually only minimizes the total length of the path, ignores the priority difference between residents, and is difficult to accurately serve high-risk groups. Q-learning can be used to build a "state-action-reward" system, and the resident portrait score, service frequency and spatial distance are jointly included in the learning framework, so as to realize the intelligentization and value driving of path scheduling.

[0058] For example, in a certain community pilot, residents A, B, C and D are located in different units, among which A and D have the highest psychological risk score, but the distance on the map is relatively far. The traditional greedy algorithm selects the order of B→C→D→A, the path is short but the high-risk objects are not covered early; and the strategy trained by Q-learning visits A and D first, although the path length increases by 8%, but the average response time of high-risk objects is shortened by 42%, the service quality is improved significantly, which proves the practical value and strategy superiority of reinforcement learning method in this scene.

[0059] Step S30: performing a visiting process according to the optimized visiting path; converting the conversation content of the visiting process into conversation text data in real time through voice recognition; calculating a resident emotional stress index based on the conversation text data combined with a BERT semantic analysis method;

[0060] It should be noted that in step S30, the visiting process is performed according to the optimized visiting path; the conversation content of the visiting process is converted into conversation text data in real time through voice recognition; and a resident emotional stress index is calculated based on the conversation text data combined with a BERT semantic analysis method, which specifically includes:

[0061] Step S301: performing a visiting process according to the optimized visiting path, and synchronously recording the conversation content in the visiting process by using a mobile terminal, converting the voice stream data of the conversation content into conversation text data in real time through an integrated voice recognition engine Wav2Vec 2.0;

[0062] Step S302: inputting the conversation text data into a pre-trained Chinese combined BERT semantic analysis method for semantic embedding coding to obtain a semantic vector of each round of conversation; setting a semantic risk keyword template set, and calculating a semantic similarity score of each round of conversation by using a cosine similarity method;

[0063] Step S303: when the semantic similarity score exceeds a preset semantic similarity score threshold, performing feature extraction on the semantic vector of the round of conversation by using a convolutional neural network to obtain emotional word feature recognition, emotional intensity feature, language repetition feature, speech speed change feature and word skipping frequency feature;

[0064] Step S304: Based on the emotion word feature recognition, the emotional intensity feature, the language repetition feature, the speech speed change feature and the word jump frequency feature, the resident emotional stress index is output by weighted linear combination.

[0065] It can be understood that by combining speech-to-text, deep semantic analysis and emotion feature modeling, the potential emotional changes of residents in conversation, such as anxiety, depression and resistance, can be automatically mined, and a quantitative index, the resident emotional stress index PI, is output, realizing the effective conversion of "unstructured conversation -> structured risk judgment", and significantly enhancing the emotional recognition ability and decision assistance ability of social workers in the visiting process.

[0066] It should be understood that the traditional visiting service record is mainly based on subjective record or coarse-grained emotional label, and lacks real-time and quantification. However, the present application can effectively identify the emotional fluctuations of "indirect language expression" such as "euphemistic refusal" and "indirect denial" through the context understanding ability of BERT model and the emotion feature extraction ability of CNN combined with the self-defined keyword template, which significantly improves the recognition accuracy of implicit risk groups.

[0067] For example, in a visiting process, the resident's expression content is: "There is really no problem, I am used to it, and I don't need to disturb you." The traditional record is only classified as "no service demand"; but in the present system, the BERT model calculates the semantic similarity with "neglected" and "no one to take care of" as 0.82, the convolutional neural network extracts the language repetition ("I am used to it" is repeated 3 times) and the word jump feature is significant, and finally calculates the resident emotional stress index as 0.81, automatically listing it as "emotional high-pressure observation object". The follow-up interview confirms that the resident has a persistent lonely emotion, which proves the effective recognition ability of the present method to implicit emotional signals.

[0068] Step S40: According to the conversation text data and the resident emotional stress index, information is structured and archived by using the graph database Neo4j, and a visiting data record unit is constructed;

[0069] It can be understood that compared with the traditional record means mainly in the form of table storage, the graph database supports the expression of multi-source data involved in the social worker's visit in the form of "node + relationship", and retains the natural connection logic between various types of information, so as to realize the visualization tracking and semantic path mining of "service process" "emotional change" "keyword trend" "risk behavior chain".

[0070] It should be understood that the traditional visit data record often faces the problems of information fragmentation, field redundancy, and difficulty in global correlation, especially in multiple visits or multi-person collaborative service process, and lacks unified data archiving standards, resulting in insufficient risk identification and trend judgment ability. The present application constructs a multi-layer relationship network of "resident-event-semantic-emotion" through the introduction of a graph database, realizes the structured organization, long-term storage and intelligent retrieval of data, and provides high-quality basic data for subsequent portrait updating and risk early warning.

[0071] For example, in one visit, the conversation text of resident B "living alone for too long, many things don't want to take care of" is automatically transcribed, and keywords such as "loneliness" and "mental fatigue" are extracted; combined with the semantic score and the resident emotional stress index 0.84, the conversation text data and the resident emotional stress index node are automatically generated and entered into the Visit path of the resident. When the resident emotional stress index is higher than 0.75 for 3 consecutive visits, abnormal trends can be found through path query, and the "psychological counseling" theme frequency is automatically added to the "emotional risk residents" through clustering analysis in the graph database, and early intervention is successfully triggered.

[0072] Step S50: outputting an emotional risk resident set according to the visit data record unit, and updating a resident portrait vector according to the visit data record unit.

[0073] It should be noted that in step S50, the steps of outputting an emotional risk resident set according to the visit data record unit and updating a resident portrait vector according to the visit data record unit specifically include:

[0074] Step S501: extracting the sequence data of the emotional stress index of the resident and the sequence data of the conversation text data from the visit data record unit;

[0075] Step S502: performing risk feature extraction on the sequence data of the emotional stress index of the resident and the sequence data of the conversation text data to obtain a risk feature vector, the risk feature vector including an emotional feature sub-vector, a text feature sub-vector and a service feedback feature sub-vector;

[0076] Step S503: inputting the risk feature vector into a lightweight classifier, the lightweight classifier outputting a resident emotional risk level, and outputting residents with a resident emotional risk level greater than a preset resident emotional risk level threshold as an emotional risk resident set;

[0077] Step S504: finally updating the resident portrait vector according to the visit data record unit.

[0078] It can be understood that the step can quantitatively identify the implicit psychological risk population without the participation of doctors or psychological experts by constructing the risk identification feature structure of "multi-round dialogue + emotional change + service response", and provides more accurate basic information support for the next round of social worker scheduling by adaptively updating the portrait vector, forming a data-driven closed-loop feedback mechanism.

[0079] It should be understood that traditional portrait updating mainly depends on static attributes (such as age, family structure), and the updating frequency is low, the granularity is coarse, and the strategy is fixed, which is difficult to reflect the dynamic changes of the state of residents. The present application regards the state of residents as a time dynamic variable, and performs fine-grained modeling combined with continuous visit data, especially by a lightweight model to output a risk level in real time, which has the advantages of strong real-time performance, easy deployment of model, timely feedback, and is suitable for high-frequency execution scenarios of daily social worker services.

[0080] For example, in three visits, resident C shows repeated expressions such as "annoyed", "helpless", "don't want to talk", and the emotional stress index sequence is 0.62, 0.75, and 0.84. The model identifies that the emotional fluctuation is severe, the service request is high but not closed loop, and the risk score is 0.89, which is higher than the set threshold 0.8. Accordingly, the system lists it as an "emotional high-risk population" and enhances the "psychological counseling intention" and "loneliness tendency" dimensions in the portrait. Subsequent social workers prefer to arrange psychological service visits, intervene in potential crises in advance, and show the effectiveness and practicality of the method in identifying implicit groups.

[0081] Embodiment two: In addition, the present application provides a visit service implementation system based on AI social workers, which adopts the visit service implementation method based on AI social workers in the above embodiment, and can solve the technical problem of visit service implementation based on AI social workers. Compared with the prior art, the visit service implementation system based on AI social workers provided by the present application has the same beneficial effects as the visit service implementation method based on AI social workers provided by the above embodiment, and other technical features in the visit service implementation system based on AI social workers are the same as the features disclosed in the above embodiment method. Here, it is not repeated.

[0082] Embodiment three: The present application provides a visit service implementation device based on AI social workers, please refer to Figure 2An AI social work-based visit service implementation device includes at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform an AI social work-based visit service implementation method of Embodiment 1. The AI social work-based visit service implementation device according to an embodiment of the present disclosure can include, but is not limited to, a mobile terminal such as a mobile phone, a notebook, a digital broadcasting receiver, a PDA (Personal Digital Assistant), a PAD (Portable Application Description), a PMP (Portable Media Player), a car terminal (e.g., a car navigation terminal), and the like, and a stationary terminal such as a digital TV, a desktop computer, and the like. The AI social work-based visit service implementation device is only an example, and should not impose any limitation on the function and use range of the embodiments of the present disclosure. The AI social work-based visit service implementation device can include a processing device 1001 (e.g., a central processing unit, a graphic processing unit, or the like) that can perform various appropriate actions and processes according to a program stored in a ROM (ReadOnly Memory) 1002 or a program loaded from a storage device 1003 to a RAM (Random Access Memory) 1004. In the RAM 1004, various programs and data required for the operation of the AI social work-based visit service implementation device are also stored. The processing device 1001, the ROM 1002, and the RAM 1004 are connected to each other through a bus 1005. An I / O (Input / Output) interface 1006 is also connected to the bus. In general, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, and the like; an output device 1008 including, for example, an LCD (Liquid Crystal Display), a speaker, a vibrator, and the like; the storage device 1003 including, for example, a magnetic tape, a hard disk, and the like; and a communication device 1009. The communication device 1009 can allow the AI social work-based visit service implementation device to communicate with other devices wirelessly or by wire to exchange data. Although the AI social work-based visit service implementation device having various systems is illustrated in the drawing, it should be understood that all of the illustrated systems are not required to be implemented or provided. More or less systems can be alternatively implemented or provided.

[0083] Embodiment Four: The present application also provides a computer program product comprising a computer program which, when executed by a processor, implements the steps of an AI social worker-based visit service implementation method as described above. The computer program product provided by the present application can solve the technical problem of AI social worker-based visit service implementation. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as those of the AI social worker-based visit service implementation method provided by the above-described embodiments, and are not described here in detail.

[0084] In particular, according to the embodiments disclosed by the present application, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, the embodiments disclosed by the present application include a computer program product comprising a computer program carried on a computer readable medium, the computer program comprising program code for executing the method shown in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by a processing device 1001, the above-mentioned functions defined in the method of the embodiments of the present application are executed.

[0085] It should be understood that various parts of the present application can be realized by hardware, software, firmware, or a combination thereof. In the description of the above-described embodiments, specific features, structures, materials or characteristics can be combined in any appropriate manner in any one or more embodiments or examples.

[0086] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application also intends to include these modifications and variations.

Claims

1. An AI-based social work visit service implementation method, characterized by, The method comprises: Step S10: Collecting basic attribute data, historical service record data and neighborhood interaction relationship data of community residents and establishing a resident portrait vector; a graph attention network model is constructed and trained in advance, the resident portrait vector is input into the graph attention network model, and a visiting priority score of the resident is output; Step S20: generating a high-priority resident set according to the visiting priority score; obtaining an actual spatial distance, and calculating an optimized visiting path by using a Q-learning reinforcement algorithm according to the high-priority resident set and the actual spatial distance; Wherein, according to the visiting priority score, a high-priority resident set is generated; an actual spatial distance is obtained, and an optimized visiting path is calculated by using a Q-learning reinforcement algorithm according to the high-priority resident set and the actual spatial distance; the step specifically comprises: The preset visiting priority score threshold is used to screen out a high-priority resident set with a visiting priority score higher than the visiting priority score threshold ; Defining a state space S of the Q-learning reinforcement algorithm, the state space S including a resident path sequence that has completed visiting, a current social worker position and a remaining visitable resident set; defining an action space A of the Q-learning reinforcement algorithm, the action space A including a set of next target residents selected from the current position; obtaining an actual spatial distance, and designing a reward function R of the Q-learning reinforcement algorithm according to the actual spatial distance and the high-priority resident set; According to the state space S, the action space A and the reward function R, a Q-learning reinforcement learning algorithm is used to perform iterative updating, and an optimized visiting path is output Wherein, in the training process of the graph attention network model, a neighborhood interaction relationship graph structure consistency constraint is introduced to construct a loss function, and the loss function adopts a formula as follows: ； wherein, is a loss function; is the total number of residents in the training set during the training process; is the real priority label of the resident i; is the predicted priority label of the resident i predicted by the graph attention network model; is a neighborhood consistency regularization factor, used to control the smoothing of the distribution under the guidance of the graph structure; is a set of adjacent edges in the neighborhood interaction relationship data; is the real priority label of the resident j; Step S30: performing a visiting process according to the optimized visiting path; converting the conversation content of the visiting process into conversation text data in real time through voice recognition; and calculating a resident emotional stress index based on the conversation text data and a BERT semantic analysis method; Step S40: information is structured and archived by using a graph database Neo4j according to the conversation text data and the resident emotional stress index, and a visiting data record unit is constructed; Step S50: outputting an emotional risk resident set according to the visiting data record unit, and updating the resident portrait vector according to the visiting data record unit. 2.The AI-based social work visit service implementation method of claim 1, wherein In step S10, the basic attribute data includes age data, gender data, residence state, marital status and health status; the historical service record data includes service frequency, problem closure rate and service satisfaction; and the neighborhood interaction relationship data has a graph structure, and includes a spatial adjacency relationship of residents, a community event cooperation relationship, a community risk transmission factor and a neighborhood node clustering coefficient. 3.The AI-based social work visit service implementation method of claim 1, wherein In step S10, the structure of the graph attention network model specifically comprises: an input layer for receiving the resident portrait vector; two hidden layers including a first hidden layer of 64 dimensions and a second hidden layer of 32 dimensions; a multi-head graph attention layer for performing attention weighted aggregation on each resident portrait vector and the resident portrait vectors of neighbors thereof; and a resident priority regression output layer adopting a Sigmoid activation function, for outputting a visiting priority score of the resident. 4.The AI-based social work visit service implementation method of claim 1, wherein In step S30, the visiting process is performed according to the optimized visiting path; the conversation content of the visiting process is converted into conversation text data in real time through voice recognition; and the resident emotional stress index is calculated based on the conversation text data in combination with the BERT semantic analysis method, specifically including: Step S301: The visiting process is performed according to the optimized visiting path, and the conversation content in the visiting process is recorded in real time by using a mobile terminal, the voice stream data of the conversation content is converted into text data in real time by using an integrated voice recognition engine Wav2Vec 2.0, and the text data is obtained; Step S302: The text data is input into a pre-trained Chinese combined BERT semantic analysis method for semantic embedding coding to obtain a semantic vector of each round of conversation; a set of semantic risk keyword templates is set, and the semantic similarity score of each round of conversation is calculated by using a cosine similarity method; Step S303: When the semantic similarity score exceeds a preset semantic similarity score threshold, the semantic vector of the round of conversation is feature-extracted by using a convolutional neural network to obtain an emotional word feature recognition, a sentiment intensity feature, a language repetition feature, a speech rate change feature and a word skipping frequency feature; Step S304: The resident emotional stress index is output by using a weighted linear combination based on the emotional word feature recognition, the sentiment intensity feature, the language repetition feature, the speech rate change feature and the word skipping frequency feature. 5.The AI social work-based visit service implementation method of claim 1, wherein, In step S50, the emotional risk resident set is output according to the visiting data recording unit, and the resident portrait vector is updated according to the visiting data recording unit, specifically including: Step S501: The sequence data of the emotional stress index and the sequence data of the conversation text data of the resident are extracted from the visiting data recording unit; Step S502: The sequence data of the emotional stress index and the sequence data of the conversation text data of the resident are risk feature-extracted to obtain a risk feature vector, and the risk feature vector includes an emotional feature sub-vector, a text feature sub-vector and a service feedback feature sub-vector; Step S503: The risk feature vector is input into a lightweight classifier, the lightweight classifier outputs a resident emotional risk level, and the residents whose resident emotional risk levels are greater than a preset resident emotional risk level threshold are output as the emotional risk resident set; Step S504: Finally, the resident portrait vector is updated according to the visiting data recording unit.

6. An AI-based social worker visit service implementation system, applied to the AI-based social worker visit service implementation method of any one of claims 1-5, characterized in that, The visiting service implementation system based on AI social work includes: The portrait construction module is used for collecting the basic attribute data, the historical service record data and the neighborhood interaction relationship data of the community residents and establishing the resident portrait vector; a graph attention network model is pre-constructed and trained, the resident portrait vector is input into the graph attention network model, and the visiting priority score of the resident is output; The path scheduling module is used for generating a high-priority resident set according to the visiting priority score; an actual spatial distance is obtained, and the Q-learning reinforcement algorithm is used to calculate the optimized visiting path according to the high-priority resident set and the actual spatial distance; The high-priority resident set is generated according to the visiting priority score; the actual spatial distance is obtained, and the Q-learning reinforcement algorithm is used to calculate the optimized visiting path according to the high-priority resident set and the actual spatial distance, specifically including: a preset visiting priority score threshold, screening out a high-priority resident set with a visiting priority score higher than the visiting priority score threshold ; The state space S of the Q-learning reinforcement algorithm is defined, and the state space S includes the resident path sequence of the completed visit, the current social worker position and the remaining visitable resident set; the action space A of the Q-learning reinforcement algorithm is defined, and the action space A includes a set of next target residents selected from the current position; the actual spatial distance is obtained, and the reward function R of the Q-learning reinforcement algorithm is designed according to the actual spatial distance and the high-priority resident set; The Q-learning reinforcement learning algorithm is used to perform iterative update according to the state space S, the action space A and the reward function R, and the optimized visiting path is output In the training process of the graph attention network model, the neighborhood interaction relationship graph structure consistency constraint is introduced to construct a loss function, and the loss function adopts the formula: ； wherein, is a loss function; is the total number of residents in the training set during the training process; is the true priority label of the resident i; is the predicted priority label of the resident i predicted by the graph attention network model; is a neighborhood consistency regularization factor for controlling the distribution smoothing under the guidance of the graph structure; is a set of adjacent edges in the neighborhood interaction relationship data; is the true priority label of the resident j; The visiting execution and emotion analysis module is used to execute the visiting process according to the optimized visiting path; the conversation content of the visiting process is converted into conversation text data in real time through voice recognition; the resident emotional stress index is calculated based on the conversation text data and the BERT semantic analysis method; The structured archiving module is used to structure and archive information using the graph database Neo4j according to the conversation text data and the resident emotional stress index, and to construct a visiting data record unit; The risk identification and portrait updating module is used to output an emotional risk resident set according to the visiting data record unit, and to update the resident portrait vector according to the visiting data record unit.

7. An AI-based social work visit service implementation device, characterized by, The AI social worker-based visiting service implementation device includes a memory, a processor, and an AI social worker-based visiting service implementation program stored on the memory and executable on the processor, and the AI social worker-based visiting service implementation program is executed by the processor to implement the AI social worker-based visiting service implementation method of any one of claims 1-5.

8. A computer program product, characterised in that, The computer program product includes an AI social worker-based visiting service implementation program, and the AI social worker-based visiting service implementation program is executed by the processor to implement the AI social worker-based visiting service implementation method of any one of claims 1-5.

Images