CN114862006A

CN114862006A - Social work service scheme automatic generation method and device based on artificial intelligence

Info

Publication number: CN114862006A
Application number: CN202210462053.XA
Authority: CN
Inventors: 赵文君; 刘柳; 李军
Original assignee: Nanjing University
Current assignee: Nanjing University
Priority date: 2022-04-28
Filing date: 2022-04-28
Publication date: 2022-08-05

Abstract

The invention discloses a social work service scheme automatic generation method and device based on artificial intelligence. The social work service scheme automatic generation method based on artificial intelligence comprises the following steps: acquiring case master basic information and background text description; generating a case main label in a semi-automatic mode; identifying problems and requirements of case owners by artificial intelligence; generating a target, a theory and a plan of a service scheme by artificial intelligence; the artificial intelligence evaluates the effectiveness of the service plan. The invention introduces the artificial intelligence technology into the social work service field, can help social workers to recommend proper service schemes and evaluate the effectiveness of the service schemes, can effectively improve the efficiency and the scientificity of the social work service, and has wide application prospect in the fields of social worker talent culture and practical work.

Description

Social work service scheme automatic generation method and device based on artificial intelligence

Technical Field

The invention relates to the field of artificial intelligence social work service; in particular to a method for assisting social workers to generate a proper service scheme and evaluating the effectiveness of the service scheme through an artificial intelligence technology; in particular to a social work service scheme automatic generation method and a device based on artificial intelligence.

Background

Artificial Intelligence (AI) is a new technical science to study and develop theories, methods, techniques and application systems for simulating, extending and expanding human Intelligence. The field research comprises robots, knowledge maps, voice recognition, image recognition, natural language processing, machine learning and the like.

Natural Language Processing (NLP) is an important direction in the fields of computer science and artificial intelligence. It studies various theories and methods that enable efficient communication between humans and computers using natural language. The natural language processing is mainly applied to the aspects of machine translation, public opinion monitoring, automatic summarization, viewpoint extraction, text classification, question answering, text semantic comparison, voice recognition, Chinese OCR and the like.

Machine Learning (ML) is a science of artificial intelligence, and the main research object in this field is artificial intelligence, particularly how to improve the performance of a specific algorithm in empirical Learning. Common machine learning algorithms include linear regression, logistic regression, support vector machines, clustering, decision trees, random forests, artificial neural networks, deep learning, and the like.

The recommendation system is an important application field of artificial intelligence, and can deduce the articles which a user may like through some mathematical algorithms. The recommendation algorithms can be roughly divided into three categories: content-based recommendation algorithms, collaborative filtering recommendation algorithms, and knowledge-based recommendation algorithms. The Collaborative Filtering recommendation algorithm is most common, and can be divided into a User-based Collaborative Filtering algorithm (User-based Collaborative Filtering), an Item-based Collaborative Filtering algorithm (Item-based Collaborative Filtering), and a Model-based Collaborative Filtering algorithm (Model-based Collaborative Filtering).

Social work is to take on the view of the value of the interest, based on scientific knowledge, and apply scientific professional methods to help the needed difficult groups, solve the problem of life dilemma, and assist individuals and social environments to better adapt to each other in professional activities. Social work is characterized by providing scientifically effective services to people in need, particularly to difficult groups. The social work is centered on the needs of helped people and takes scientific assistant skills as means to achieve the effectiveness of the helpers.

The social work service specific method comprises individual case work, group work and community work, and generally relates to six processes of case receiving, estimation, planning, implementation, evaluation and case settlement. Due to the particularity of social work services themselves and the scarcity of the relevant talents, the application of information technology in the field of social work has not attracted sufficient attention for a long time. The formulation of the service scheme in the current social work and practice field generally depends on the personal experience of social workers, and the successful service experience is difficult to summarize and popularize, so that the efficiency and the scientificity of the social work service scheme are difficult to guarantee. At present, the artificial intelligence technology is not widely applied to the fields of social work education, scientific research and practice.

The invention introduces the artificial intelligence technology into the social work service field, can help social workers to recommend proper service schemes and evaluate the effectiveness of the service schemes, can effectively improve the efficiency and the scientificity of the social work service, and has wide application prospect in the fields of social worker talent culture and practical work.

Disclosure of Invention

The invention solves the problems of efficiency and scientificity of the social work service scheme to a certain extent.

Therefore, the first purpose of the invention is to provide an artificial intelligence-based social work service scheme automatic generation method. The method comprises the following steps:

the method comprises the steps of firstly, acquiring basic information and a background description text of a case;

secondly, generating a case main label in a semi-automatic mode;

thirdly, identifying problems and requirements of case owners by artificial intelligence;

fourthly, generating a target, a theory and a plan of the service scheme by artificial intelligence;

and fifthly, evaluating the effect of the service scheme by artificial intelligence.

The second purpose of the invention is to provide an automatic social work service scheme generation device based on artificial intelligence. The device comprises two subsystems: an artificial intelligence model training subsystem and a social work service management subsystem.

The social work service scheme automatic generation method based on artificial intelligence can generate a scheme main label based on natural language processing technology, machine learning algorithm and manual auditing semi-automation according to the description text of the scheme main background. The natural language processing technology is mainly used for performing word segmentation and syntactic analysis on a text and extracting keywords, phrases and events in the text; the machine learning technology is mainly used for establishing a classification algorithm model of keywords, phrases and events and corresponding labels and automatically generating case main labels; the manual review is mainly used for reviewing and correcting the accuracy of the automatically generated label so as to guarantee the accuracy of the label used for the case master.

A case Tag (Tag) is a generalization of the meaning implied by a particular sentence in a descriptive text. By artificially tagging each Sentence (sequence) describing the text, a Sentence-Tag training Set (STS) can be created; each sentence is composed of a group of meaningful keywords (keywords), phrases (Phase) or events (events), wherein the keywords are mainly nouns, verbs, adjectives and other meaningful important words, the phrases are mainly composed of a plurality of continuous nouns, verbs and adjectives, and the events can be represented by a three-tuple of a main subject and a predicate. Therefore, keywords, phrases and events of each sentence can be extracted based on the natural language processing technology, and each keyword, phrase or event can be regarded as one Code, so that a Code-Tag Set (CTS) can be established based on the sentence-Tag training Set. There are actually three subsets of CTS, namely, Keyword-Tag training Set (KTS), phrase-Tag training Set (PTS) and Event-Tag training Set (ETS).

Since there may be a negative determination in a sentence, the sentence must be parsed, and if the scope of the negative word is valid for the keyword, phrase, and event at the time of the parsing, the keyword, phrase, and event are identified as a negative state, and the default condition is a positive state.

Based on any subset of CTS, a recognition model of the tag can be established by a classification algorithm. And finding the optimal algorithm as the label classification algorithm of the CTS subset through training of a training set and a testing set. For each tag, there are three classification algorithms, namely a KTS-based classification algorithm, a PTS-based classification algorithm, and an ETS-based classification algorithm. Because each label is actually associated with some specific codes, all related codes of each label can be used as the characteristic codes of the label according to the corresponding relation between the codes and the labels in the training set, so that the index relation between the codes and the labels can be quickly established, namely the labels possibly corresponding to the labels can be quickly retrieved through the codes.

The automatic generation of the case owner label is to perform label identification on each sentence of the case owner description text. After each sentence is extracted and encoded, a potential tag set of the sentence can be obtained according to the tag set which can possibly correspond to each encoding. For each label in the set, three classification algorithms for that label are computed one by one. The results of the three classification algorithms for each label can be used in a variety of comprehensive ways to identify the final label classification.

Since a keyword typically has multiple synonyms, it is unlikely that all synonyms are contained in the sample training set. To improve the efficiency of tag identification, the CTS needs to introduce a synonym table to extend keywords, phrases, and events. The synonym table maintains a set of synonyms for commonly used keywords. The synonym table can be constructed manually or based on a word vector similarity algorithm.

The automatically identified label needs to be checked and confirmed manually so as to ensure the accuracy of the label; due to the lack of training set samples, in some cases, tags may be manually deleted or new tags appended.

According to the social work service scheme automatic generation method based on artificial intelligence, after the background information of the case owner is labeled, the problems and requirements of the case owner can be identified based on an artificial intelligence algorithm. The problem of the case owner can be summarized into the types of survival problem, development problem, protection problem, relation problem, psychological problem, health problem, behavior problem, learning problem, employment problem and the like, and the problems can be enumerated according to the existing experience, so that the problem identification of the case owner is a multi-classification problem, can be realized through an artificial intelligent classification algorithm, such as a decision tree algorithm, a multi-class perceptron algorithm and the like, and finally, the accuracy of problem identification is confirmed manually by social workers. The requirement of the case master is based on corresponding individual requirements derived from basic information of the case master, background tags of the case master and problems of the case master, the basic information of the case master and the case master with similar background tags possibly have greater similarity in service requirements, but possibly have differences, so that the requirement identification of the case master is a recommendation problem, potential requirements can be recommended to the case master with similar background of the case master by adopting recommendation algorithms such as collaborative filtering and the like, and finally, social workers can manually decide which specific requirements to reserve or newly add, so that the social workers can obtain more comprehensive cognition on the requirements of the case master, and the efficiency, accuracy and comprehensiveness of requirement identification are also improved.

After the problems and requirements of a case owner are identified, the social work service scheme automatic generation method based on artificial intelligence can generate the target, theory and plan of the service scheme based on an artificial intelligence algorithm. The target of the service scheme usually corresponds to the requirement, i.e. the target is to meet a certain requirement of the case owner, and of course, the target can be a one-to-many or many-to-many relationship. The target usually also considers the background and the problem of the case owner, therefore, the generation of the target can also be regarded as a recommendation problem, namely, the case owners with similar background, problem and requirement, the targets of the service schemes are also similar, and the target generation algorithm can also use a recommendation algorithm such as collaborative filtering. Similarly, the theory and plan of the service scheme are also a recommendation problem, namely a case master with similar background, problem, demand and target, and the theoretical basis and implementation plan of the service scheme are also similar. The theoretical basis of the service plan is a subset of theories, and related theories can be recommended from the whole set of theories. The planning of the service plan is usually a task queue, each task comprises time, intervention action type and service mode, and the number of the task queue can be determined according to the average number of tasks planned by all cases with similar case background, problem, demand and target in the existing case training set. An intervention action type is also a set of types, and a service plan may take a variety of intervention actions. Thus, the goals, theories, and types of intervention actions in the plan for the service plan may all be generated based on the recommendation algorithm.

To enhance the user experience, the service plan may generate a task queue with an order based on statistical characteristics of the type of intervention actions recommended by the algorithm. The method comprises the following implementation steps:

firstly, acquiring a similar case set;

secondly, acquiring the characteristic distribution of the intervention action;

thirdly, determining the type and the number of intervention actions;

fourthly, determining the sequence of the intervention actions;

and fifthly, generating a detailed service plan.

After the goal, the theory and the plan of the service scheme are generated, the effect of the service scheme can be evaluated based on an artificial intelligence algorithm. The effect of the service scheme is usually scored using five points, with a very good effect of 5 points and a very poor effect of 1 point. Therefore, the evaluation algorithm of the service scheme can predict the score by adopting a BP neural network algorithm or other neural network algorithms based on the existing case training set, the other neural network algorithms comprise a Radial Basis Function (RBF) neural network, a perceptron neural network, a linear neural network and the like, and a most appropriate algorithm can be selected according to the test effect of the case training set in practical application.

The artificial intelligence model training subsystem of the automatic social work service scheme generating device based on artificial intelligence comprises: the system comprises a case data input module, a case data processing module and a case data training module.

The case data entry module can manually or semi-automatically enter the complete data structure of the social work service case into the system, wherein the complete structure of the case data comprises case main basic information, case main background text description, case main problems, case main requirements, service targets, service theories, service plans, service implementation processes, service evaluation and service end plans.

The case data processing module mainly extracts keywords, phrases and event codes of each sentence in the text for case main background word descriptions based on a natural language processing technology, and classifies the sentences to corresponding case main labels based on a pre-trained label recognition model and a multi-label classification algorithm. In addition, the data processing module also supports manual or semi-automatic labeling representation of case master requirements and service targets, and labeling of case master requirements and service targets can be assisted through phrase extraction technology.

The case data training module mainly takes the structured data of the case data processing module as a data set, and trains based on a machine learning algorithm and a recommendation algorithm to generate a corresponding artificial intelligence model. The training model mainly comprises: the system comprises a label identification model, a problem identification model, a demand identification model, a target generation model, a theoretical generation model, a plan generation model and an effect evaluation model. The problem identification model and the effect evaluation model are machine learning models, and the rest are recommendation models.

The social work service management subsystem of the automatic social work service scheme generation device based on artificial intelligence comprises: the system comprises a case receiving module, an estimation module, a planning module, an implementation module, an evaluation module and a case settlement module. The case receiving module can integrate the label identification model to carry out case owner labeling; the pre-estimation module can integrate a problem identification model and a demand identification model, and assists social workers to identify problems and demands of case owners more efficiently and accurately; the plan module can integrate a target generation model, a theoretical generation model and a plan generation model, and assists social work to more efficiently and comprehensively make targets, theories and plans of a service scheme; the implementation module provides process management for the social worker to actually carry out social work service, and comprises a plurality of social work service records; before or after each service record is developed, the implementation module can integrate the effect evaluation model to dynamically predict the effect of the service scheme in real time; after the implementation stage of the social work service is completed, the evaluation module provides satisfaction evaluation of case owners and active evaluation and summary of thinking of social workers, and compares the evaluation result with the evaluation effect predicted by the effect evaluation model; the settlement module provides follow-up arrangement after the social work service is finished.

After the social work service management subsystem completes the processes of receiving, estimating, planning, implementing, evaluating and settling the social service, the structured service overall process data can also become a case training set of the model training subsystem for automatically updating and perfecting the parameters of the model.

Drawings

FIG. 1 is a flow chart of a method for automated generation of a social work service scheme based on artificial intelligence.

Fig. 2 is a flow chart of implementing automatic labeling for the description text of the case master background.

FIG. 3 is a schematic diagram of a synonym set.

FIG. 4 is a schematic diagram of an interface automatically generated by a social work service offering.

FIG. 5 is an interface diagram of a social work service scenario effectiveness assessment.

FIG. 6 is a flow chart of generating a task queue with order based on statistical characteristics of intervention action types recommended by an algorithm when a service plan is generated.

Fig. 7 is a schematic structural diagram of an artificial intelligence-based social work service scenario automation generation apparatus.

Detailed Description

The process of the invention is described below with reference to the accompanying drawings:

referring to fig. 1, fig. 1 is a flowchart of a social work service scenario automatic generation method based on artificial intelligence.

11 obtaining case main basic information and background description text

The basic information of the case owner includes the gender, age, academic calendar, nationality, income, marital, work, etc. of the case owner, and the case owner background description text is a text for describing the condition of the case owner. This information is typically entered into the device by social workers at the pick-up desk.

For example, a paragraph of main background text in one embodiment of the present invention is described below: the physician is in charge of the multiple disabilities of the small W, male, age 19, and diagnosed as "hyperkinetic" by the physician. Parents leave the parents 10 years ago and live together with mothers all the time, the living environment of the parents is poor, the parents belong to low-security families, the mothers maintain family expenses by buying and selling fruits, and the counter often laughs or buys strangers due to the reasons of brain function damage and the like. Since no one cares about the case at home, the mother can only lock one person at home to do business. The case owner is locked at home alone, and no one talks, and jokes or shores on a balcony frequently facing a window or standing, so that the normal rest of the neighbor is seriously influenced, the neighbor complains to the residence committee many times, and the residence committee is innocent, and the case is transferred to the social worker.

12 scheme main labeling based on natural language processing technology

The text description of the case main background is inconvenient for quantitative calculation and needs to be labeled. The specific processing manner is described with reference to fig. 2.

For example, a case master context textual description in one embodiment of the invention may correspond to a case master tag as follows: hyperactivity, disability, intellectual disability, mental illness, dissimilarity to the home, economic difficulties, influencing neighbors, complaints, etc.

13 identifying problems and needs of case masters based on artificial intelligence algorithm

The problem identification is a multi-classification problem and can be realized by an artificial intelligent classification algorithm, such as a decision tree algorithm, a multi-class perceptron algorithm and the like. The labels of the case owner are used as input, each label is a variable of 0 or 1, and a two-classification algorithm can be established for each problem, so that multiple classifications of the labels are realized. One host may be involved with several problems.

Demand recognition is a recommendation problem, i.e. similar case masters and backgrounds may have similar demands, which may be implemented by a user-based collaborative filtering algorithm.

For example, chief complaints in one embodiment of the invention include hyperactivity disorder, intellectual impairment, and volitional behavior disorder, all of which fall into the broad category of health problems. Based on the analysis of a large number of similar cases, the potential needs may include increasing self-care ability, reducing security risks, increasing household income, improving neighborhood relations, and the like. Social workers may manually choose to improve self-care and reduce security risks as the main requirements.

14 generating goals, theories, and plans for a service plan based on an artificial intelligence algorithm

The generation of the target, theory and plan of the service scheme is a recommendation problem, namely, the case masters with similar backgrounds, similar problems and detailed requirements may have similar targets, theories and plans, and can be realized by a collaborative filtering algorithm based on users.

The service target is a group of tagged target lists which can be supplemented manually by social workers, the service theory is also a group of enumerable theory lists, and the service plan protects a group of enumerable intervention action types.

For example, the recommended target of the case master service scheme in one embodiment of the invention comprises the functions of improving self-care ability, improving cognition and improving behavior, the recommended theory is cognitive behavior theory and social capital theory, and the recommended intervention action type is professional consultation, service placement and resource integration. The specific schematic diagram can be seen in fig. 4.

15 evaluating the effectiveness of the service plan based on an artificial intelligence algorithm

The service plan evaluation can be based on the existing case training set, and the score can be predicted by adopting a BP neural network algorithm or other neural network algorithms.

For example, the effectiveness evaluation score for a recommended service plan for a employer in one embodiment of the present invention may be 4.8 points. The specific schematic diagram can be seen in fig. 5.

Referring to fig. 2, fig. 2 is a flow chart of implementing automatic labeling for the description text of the case main background.

The description text of the case main background is labeled, the traditional method only can be manually coded by people and usually needs to rely on social workers or researchers with professional knowledge and rich experience. This results in reduced coding efficiency, increased cost, and no guarantee of coding efficiency and reliability. In addition, the traditional text classification is generally classified only based on the distribution characteristics of the keywords, and the accuracy is difficult to improve, such as the traditional spam classification algorithm. Because the labels of the text relate to definite semantics and concept extraction, the invention provides a multiple label classification algorithm based on keywords, phrases and events at the sentence level, integrates three technical means of keyword extraction, phrase extraction and event extraction to complete the encoding of the text, trains a training set of the encoding and labels based on a machine learning algorithm, and generates a label classification model based on the keywords, a label classification model based on the phrases and a label classification model based on the events. The three models are used simultaneously, so that the accuracy and efficiency of label classification are improved; the following is a specific implementation procedure:

21 sentence segmentation

The description text of the case main background is divided into sentences first.

22 text segmentation and syntactic analysis

Performing text segmentation and syntactic analysis on each sentence, wherein the current mainstream Chinese segmentation and syntactic analysis toolkits such as ANSJ, Stanford NLP, HaNLP and the like can be used;

23 keyword extraction

After the text is divided into words, keyword extraction can be respectively carried out, and the keywords comprise nouns, verbs and adjectives.

For example, a sentence may include 3 keywords corresponding to K1, K2, and K3.

24 phrase extraction

Phrase extraction can be respectively carried out after the text is divided into words, and the phrases can be extracted based on an N-Gram algorithm.

For example, a sentence may include 3 phrases, which correspond to P1, P2, and P3, respectively.

25 event extraction

The text can be divided into words and then event extraction can be carried out, the event can be analyzed in a syntactic mode based on a syntactic tree, and subjects and objects are extracted respectively by taking verbs as predicates. Events are usually represented by triplets, which are represented by: (subject, predicate, object), the absence of an object can be expressed as: (subject, predicate,.

For example, a sentence may include 3 events, corresponding to E1, E2, and E3, respectively.

26 negative determination

Based on the syntactic analysis of 25 steps, a negative judgment in the sentence is found and all keywords, phrases and events within the influence range of the negative word are identified as negative states.

For example, if the syntax analysis result negatively determines that the scope affects K1, P1 and E1, but does not affect other codes, the sentence is coded as { { K1 ', K2, K3}, { P1', P2, P3}, { (E1 ', E2, E3) }, where K1', P1 'and E1' are the negative states of the original K1, P2 and E1, respectively.

27 machine learning classification

From the set of tags to which the encoding may correspond, a set of potential tags for a sentence may be obtained. For each label in the set, 3 classification algorithms for that label are computed one by one.

The label classification is based on a pre-trained machine learning model, and the machine learning algorithm can be a classification algorithm such as simple Bayes, linear regression, logistic regression, decision tree, BP neural network and the like. In practical application, a machine learning algorithm with the best effect can be adopted according to practical conditions.

27 comprehensive classification results

For the results of the three classification algorithms of each tag, the final tag classification can be judged in various comprehensive ways, if any algorithm result is true, the sentence is identified as the tag, a minority obeying majority voting principle can be adopted, or the result is weighted based on a weight and then comprehensively judged. In practical application, an optimal comprehensive classification result can be adopted according to practical situations.

Referring to FIG. 3, FIG. 3 is an example of a synonym table.

The invention also provides a synonym table-based coding expansion method, which can improve the efficiency of case main label identification. Specifically, the invention provides a method for constructing a synonym table based on word vector similarity.

Since the case training set cannot exhaust all keywords, phrases and events, it is necessary to extend the keywords, phrases and events through the synonym table. Synonym tables are typically augmented for specific actions, adjectives, or entity names.

For example, such as: synonyms for "care" include "care," "nursing," "care," etc., then the phrase "care needed" can be extended to "care needed," "nursing needed," "care needed," etc., where the phrases appear in the background text, all mapped to the same case master tag.

The synonym table can also be automatically constructed by constructing a word vector of each word based on analysis of massive texts (such as news texts) and calculating the similarity of the word vectors; the phrase consists of 2 or more keywords, so that the synonym table of the keywords can be suitable for the expansion of the phrase; the event triples can respectively construct a synonym table for the subject, the predicate and the object in the triples, so that the event triples can be expanded.

In extracting keywords, phrases and events in a sentence, if the keywords, the keywords in a phrase or the keywords in an event exist in a synonym table, the keywords are replaced with reference synonyms of the synonym table.

Referring to FIG. 6, FIG. 6 is a flow chart of generating a task queue with order based on statistical characteristics of types of intervention actions recommended by an algorithm when a service plan is generated.

Traditional service plan plans rely primarily on the generation of social workers by hand. The social workers make service plan for different case owners by means of personal experience and professional knowledge. The invention not only provides a specific intervention action type which is adopted by the service scheme plan and is determined based on the collaborative filtering recommendation algorithm, but also generates a task queue with a sequence based on the statistical characteristics of the intervention action type, thereby generating a complete task plan; in addition, the invention also introduces a mode of expert online questionnaire survey, and consults the suggestions of social working experts on intervention action types for specific problems of a specific case main group as a supplement to an artificial intelligence system, thereby greatly improving the efficiency and the scientificity of service plan formulation. The method comprises the following specific steps:

61 obtaining a set of similar cases

A set of cases with similar case backgrounds, problems, requirements and goals are obtained from the training set.

Obtaining 62 a feature distribution of an interventional action

The average number of service records for a similar case set and the frequency of all intervention action types are counted.

63 determining intervention action type and number

According to the frequency ranking of the intervention action types, respectively recommending experts to suggest intervention actions, most similar case main intervention actions, most common intervention actions and least common intervention actions to social workers; the social worker may manually select one or more intervention actions and determine respective service times based on a frequency ratio in the set of similar cases.

Wherein the most similar case main intervention actions, the most common intervention actions and the least common intervention actions can be obtained by statistical feature distribution of the intervention actions of the case training set; the expert suggested intervention action can be obtained based on a large-scale online expert voting mode, namely, the suggestion of the social working expert on the intervention action type is inquired about the specific problem of the specific case main group through an online questionnaire mode, and the intervention action type of the expert suggestion aiming at the specific problem of the specific case main group is obtained through a voting mode.

In practical application, the device can provide a plurality of expert suggested intervention actions, a plurality of most similar case main intervention actions, a plurality of most common intervention actions and a plurality of least common intervention actions for selection of social workers. Each intervention action type may provide a reference index, such as how much proportion of the expert recommendations each expert suggests that the intervention action may provide, how similar degree of the case owner each most similar case owner intervention action may provide, and how proportion of the intervention action type is taken by each most common and least common intervention action.

64 determining the order of intervention actions

Calculating the probability of each intervention action in the training set appearing in the first service, and determining the highest probability as the type of the intervention action of the first service; aiming at the Nth service, determining the intervention action with the highest probability of the Nth service record in the training set as the intervention action type of the Nth service; this process is repeated until all service tasks determine the type of intervention action.

65 generating a service plan detail plan

And setting time and service modes for each intervention action, and generating a detailed plan of a case master service scheme. The interval of the common tasks is week as a time unit, and the service mode is mainly offline communication; social workers may also manually supplement the specific dates, service patterns, or modify the intervention action types for each task.

Referring to fig. 7, fig. 7 is a schematic structural diagram of an automatic social work service scenario generation apparatus based on artificial intelligence.

The automatic generation device of the social work service scheme based on artificial intelligence comprises two subsystems: an artificial intelligence model training subsystem and a social work service management subsystem.

The artificial intelligence model training subsystem comprises a case data input module, a case data processing module and a case data training module, wherein the case data training module comprises seven algorithm models: the system comprises a label identification model, a problem identification model, a demand identification model, a target generation model, a theoretical generation model, a plan generation model and an effect evaluation model.

The social work service management subsystem comprises six modules: the system comprises a case receiving module, an estimation module, a planning module, an implementation module, an evaluation module and a case settlement module.

The case receiving module can integrate a label recognition model, the pre-estimation module can integrate a problem recognition model and a demand recognition model, the plan module can integrate a target generation model, a theoretical generation model and a plan generation model, the implementation module and the evaluation module can integrate an effect evaluation model, and the case receiving module can lead complete social work service case data into the artificial intelligence model training subsystem to be used as a training set of the case data training module.

Claims

1. An automatic social work service scheme generation method based on artificial intelligence is characterized by comprising the following steps:

the method comprises the following steps:

secondly, generating a case main label in a semi-automatic mode;

2. The utility model provides a social work service scheme automated generation device based on artificial intelligence which characterized in that:

the apparatus comprises two subsystems: an artificial intelligence model training subsystem and a social work service management subsystem.

3. The method of claim 1, wherein: the method can generate the case main label based on natural language processing technology, machine learning algorithm and manual auditing semi-automation according to the description text of the case main background. The natural language processing technology is mainly used for performing word segmentation and syntactic analysis on the text and extracting keywords, phrases and events in the text; the machine learning technology is mainly used for establishing a classification algorithm model of keywords, phrases and events and corresponding labels and automatically generating case main labels; the manual review is mainly used for reviewing and correcting the accuracy of the automatically generated label so as to guarantee the accuracy of the label used for the case master.

The case Tag (Tag) is a generalization of the meaning implied by a particular sentence in the descriptive text. Each sentence is composed of a Set of meaningful keywords (keywords), phrases (Phase), or events (Event), and each Keyword, phrase, or Event can be regarded as a Code (Code), so that a Code-Tag Set (CTS) can be established.

The CTS has three subsets, which are a Keyword-Tag training Set (KTS), a phrase-Tag training Set (PTS) and an Event-Tag training Set (ETS). Based on any subset of CTS, a recognition model of the tag can be established by a classification algorithm.

The automatic generation of case master tags is to perform tag recognition on each sentence of the case master description text. After each sentence is extracted and encoded, a potential tag set of the sentence can be obtained according to the tag set which can possibly correspond to each encoding. For each label in the set, three classification algorithms for that label are computed one by one. The results of the three classification algorithms for each label can be used in a variety of comprehensive ways to identify the final label classification.

4. The method of claim 3, wherein: the code can be automatically expanded through a synonym table; the construction method of the synonym table is based on a word vector similarity method for construction.

5. The method of claim 4, wherein: after the method carries out tagging on the background information of the case owner, the problems and the requirements of the case owner can be identified based on an artificial intelligence algorithm.

The problem identification is a multi-classification problem and can be realized by a classification algorithm of artificial intelligence;

the case requirement identification is a recommendation problem, and potential requirements can be recommended to case owners with similar case owner backgrounds by adopting recommendation algorithms such as collaborative filtering and the like.

6. The method of claim 5, wherein: the method may generate goals, theories, and plans for a service plan based on an artificial intelligence algorithm after identifying the problem and need of the case owner.

The generation of the service scheme target is a recommendation problem, namely a case master with similar background, problem and requirement, and the service scheme target is similar. Similarly, the theory and plan of the service scheme are also a recommendation problem, namely a case master with similar background, problem, demand and target, the theoretical basis of the service scheme is also similar, and the case master with similar background, problem, demand, target and theory is also similar to the service plan. The generation algorithms of the target, theory and plan of the service scheme can be realized by recommendation algorithms such as collaborative filtering and the like.

7. The method of claim 6, wherein: the service plan may generate a task queue having an order based on statistical characteristics of the type of intervention actions recommended by the algorithm. The method comprises the following implementation steps:

firstly, acquiring a similar case set;

secondly, acquiring the feature distribution of the intervention action;

thirdly, determining the type and the number of intervention actions;

fourthly, determining the sequence of the intervention actions;

and fifthly, generating a detailed service plan.

The determined intervention action type and the determined intervention action number can respectively recommend an expert to suggest intervention actions, most similar case main intervention actions, most common intervention actions and least common intervention actions to social workers according to the frequency ranking of the intervention action type; the social worker may manually select one or more intervention actions and determine respective service times based on a frequency ratio in the set of similar cases.

The most similar case main intervention actions, the most common intervention actions and the least common intervention actions can be obtained by statistical feature distribution of the intervention actions of a case training set; the expert suggested intervention action can be obtained based on a large-scale online expert voting mode, namely, the suggestion of the social working expert on the intervention action type is inquired about the specific problem of the specific case main group through an online questionnaire mode, and the intervention action type of the expert suggestion aiming at the specific problem of the specific case main group is obtained through a voting mode.

8. The method of claim 6, wherein: after generating the target, theory and plan of the service scheme, the method can evaluate the effect of the service scheme based on an artificial intelligence algorithm.

9. The apparatus of claim 2, wherein: the artificial intelligence model training subsystem of the device comprises: the system comprises a case data input module, a case data processing module and a case data training module.

10. The apparatus of claim 2, wherein: the social work service management subsystem comprises six modules: the system comprises a case receiving module, an estimation module, a planning module, an implementation module, an evaluation module and a case settlement module. The case receiving module can integrate a label recognition model, the pre-estimation module can integrate a problem recognition model and a demand recognition model, the plan module can integrate a target generation model, a theoretical generation model and a plan generation model, the implementation module and the evaluation module can integrate an effect evaluation model, and the case receiving module can lead complete social work service case data into the artificial intelligence model training subsystem to serve as a training set of the case data training module.

11. The apparatus of claim 9, wherein: the case data training module mainly takes the structured data of the case data processing module as a data set, and trains based on a machine learning algorithm and a recommendation algorithm to generate a corresponding artificial intelligence model. The training model mainly comprises: the system comprises a label identification model, a problem identification model, a demand identification model, a target generation model, a theoretical generation model, a plan generation model and an effect evaluation model. The problem identification model and the effect evaluation model are machine learning models, and the rest are recommendation models.