A method, computer program product, and system for generating case supervision based on causal models and knowledge graphs

Abstract

This invention provides a method, computer program product, and system for generating psychological counseling case supervision based on causal models and knowledge graphs. The method includes: acquiring structured entity fields converted from psychological counseling dialogue text; querying a preset psychological expert corpus knowledge graph based on the structured entity fields to obtain the causal chain path and probability weight of the current intervention technique pointing to the current client's state; using the current client's state as the target, retrieving a new intervention technique that points to the target and has a positive effect from the psychological expert corpus knowledge graph; keeping the current symptoms and current risk factors unchanged, replacing the current intervention technique with the new intervention technique, and simulating and deducing the new probability weight of the new intervention technique pointing to the current client's state; comparing the deduced new probability weight with the retrieved probability weight, and if the deduced new probability weight is higher, converting its corresponding causal chain path into psychological counseling supervision text output.

Description

Technical Field

[0001] This invention relates to the field of psychological counseling data processing technology, and in particular to a case supervision generation method, computer program product and system based on causal models and knowledge graphs. Background Technology

[0002] In the field of psychological counseling, therapists need targeted feedback and supervision from supervisors when facing difficulties in handling individual cases. Traditional supervision relies on real people and usually adopts a one-to-many supervision model with one supervisor for multiple therapists, which has problems such as untimely response and scarcity of supervision resources.

[0003] Currently, AI-assisted psychological supervision technology has emerged in the industry. It uses language models to perform vectorized similarity matching between psychological counseling dialogues and pre-set expert knowledge, automatically generating and providing professional supervisory opinions, and can respond promptly to the supervision needs of multiple psychological counselors. However, this technology's vectorized similarity matching method only considers the matching degree of a single round of dialogue, ignoring the causal logic of changes in the client's state caused by intervention techniques. This can easily lead to misleading supervisory opinions when facing complex psychological counseling scenarios. Summary of the Invention

[0004] The technical problem this invention aims to solve is how to combine the causal logic of changes in the state of clients undergoing psychological counseling through intervention techniques to conduct psychological counseling supervision.

[0005] To address the aforementioned technical problems, this invention provides a method for generating psychological counseling case supervision based on causal models and knowledge graphs, comprising the following steps implemented using a structured causal model: S1. Obtain structured entity fields converted from psychological counseling dialogue text, wherein the structured entity fields include at least: symptoms, intervention techniques, client status, and risk factors; S2. Based on the structured entity fields, query the preset psychological expert corpus knowledge graph to obtain the causal chain path and its probability weight of the current intervention technology pointing to the current visitor's state; S3. Using the current state of the visitor as the target, a new intervention technique that points to the target and has a positive effect is retrieved from the knowledge graph of the psychological expert corpus; S4. Keeping the current symptoms and current risk factors unchanged, replace the current intervention technique with the new intervention technique, and simulate and deduce the new probability weights of the new intervention technique pointing to the current state of the visitor in a structured causal model; S5. Compare the new probability weights derived in step S4 with the probability weights retrieved in step S2. If the new probability weights derived in step S4 are higher, then the causal chain path of the corresponding new intervention technology pointing to the current state of the client is converted into psychological counseling supervision text output.

[0006] Furthermore, in step S5, if the probability weight obtained in step S2 is higher, then the causal chain path of the current intervention technology pointing to the current visitor's state is transformed into psychological counseling supervision text output.

[0007] Furthermore, in step S3, if multiple new intervention techniques are obtained through reverse retrieval, the multiple new intervention techniques are sorted according to the probability weight of the causal chain path pointing to the target by each new intervention technique to generate a candidate intervention list; in step S4, the current intervention technique is replaced one by one with the multiple new intervention techniques in the candidate intervention list; in step S5, the causal chain path pointing to the current client's state by the new intervention technique with the highest probability weight is converted into psychological counseling supervision text output.

[0008] Further, in step S2, the probability weights are obtained by calculating the average causal effect based on the desensitized psychological counseling records. Specifically, natural language processing technology is used to calculate the emotional score of each dialogue in the psychological counseling records as a confounding variable for the quality of the counselor-client relationship. While keeping the confounding variable unchanged, the difference between the probability of the client's emotional improvement when using a specific intervention technique and when not using the specific intervention technique is compared as the average causal effect.

[0009] Furthermore, in step S1, a large language model is specifically used as a semantic parser to convert psychological counseling dialogue text into structured entity fields.

[0010] Furthermore, in step S5, a large language model is specifically used as a natural language generator to realize the transformation of the new causal chain path and its probability weights that point the new intervention technology to the current state of the client into psychological counseling supervision text output.

[0011] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps in the above-described method for generating psychological counseling case supervision.

[0012] The present invention also provides a psychological counseling case supervision generation system based on causal model and knowledge graph, including a server, the server including a processor and a memory connected to each other, the memory storing the above-mentioned computer program product.

[0013] This invention offers the following advantages: While maintaining the current symptoms and risk factors, it replaces the current intervention technique with a new one, simulates and extrapolates the new probability weights of the new intervention technique affecting the current client's state, and when the new probability weights are higher, transforms the corresponding causal chain path and the new probability weights into a psychological counseling supervision text output. This output psychological counseling supervision text incorporates the causal logic of how the intervention technique leads to changes in the client's state, providing targeted intervention supervision opinions in complex psychological counseling scenarios. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating a method for generating psychological counseling case supervision based on causal models and knowledge graphs.

[0015] Figure 2 This is a schematic diagram of the causal ladder theory. Detailed Implementation

[0016] The present invention will be further described in detail below with reference to specific embodiments.

[0017] This embodiment provides a psychological counseling case supervision generation system based on causal models and knowledge graphs, including a server. The server includes a processor and a memory interconnected. The memory stores computer program products, which include computer programs. When executed by the processor, the computer programs implement... Figure 1 The steps in the psychological counseling case supervision generation method based on causal model and knowledge graph shown include the following steps S1, S2, S3, S4 and S5 implemented using the structured causal model (SCM).

[0018] S1. Obtain the structured entity fields converted from the psychological counseling dialogue text. The structured entity fields include at least: symptoms, intervention techniques, client status, and risk factors.

[0019] When counselors face difficulties in handling individual cases during psychological counseling, they can input the counseling dialogue of that case into this system for targeted feedback and supervision. After receiving the counseling dialogue, the system uses a Large Language Model (LLM) as a semantic parser to convert the counseling dialogue text into structured entity fields and input them into a structured causal model. In this way, the structured causal model obtains the structured entity fields converted from the counseling dialogue text. The structured entity fields include at least: symptoms (such as "feeling hopeless", "anxiety attack"), intervention techniques (such as "emotional support", "crisis assessment", "establishing a safety contract"), client status (such as "improved mood", "resistance"), and risk factors (such as "suicidal tendencies", "self-harming behavior").

[0020] S2. Based on the structured entity fields, query the preset knowledge graph of psychological expert corpus to obtain the causal chain path and its probability weight of the current intervention technology pointing to the current state of the visitor.

[0021] After obtaining the structured entity fields, they are input into the preset psychological expert corpus knowledge graph. Based on the structured entity fields, the preset psychological expert corpus knowledge graph is queried to obtain the causal chain path and probability weight of the current intervention technology pointing to the current visitor's state in the structured entity fields.

[0022] In this embodiment, the knowledge graph of psychological expert corpus is specifically designed for the field of psychological counseling and includes four core entities: symptoms (such as "feeling of hopelessness" and "anxiety attack"), intervention techniques (such as "emotional support," "crisis assessment," and "establishing a safety contract"), client states (such as "improved mood" and "resistance"), and risk factors (such as "suicidal tendencies" and "self-harming behavior"). Entities are connected through logical relationships (such as "applicable" and "prohibited") and probabilistic causal chains. The probabilistic causal chain contains the causal path from the intervention technique to the client's state and its probability weight, representing the probability weight of the intervention technique leading to a change in the client's state.

[0023] The probability weights are calculated based on the average causal effect from desensitized psychological counseling records (counseling process, client performance, counselor judgment and intervention). Specifically, natural language processing technology is used to calculate the emotional score (between -1 and 1) of each dialogue in the counseling records. This emotional score is used as a confounding variable for the quality of the counselor-client relationship. While keeping this confounding variable constant, the difference in the probability of client emotional improvement when using a specific intervention technique versus not using it is compared. This difference is taken as the average causal effect of the specific intervention technique on the client's state change. The calculated probability weights are then stored in a psychological expert corpus knowledge graph.

[0024] S3. Using the current state of the visitor as the target, retrieve new intervention techniques that point to the target and have a positive effect from the knowledge graph of psychological experts' corpus.

[0025] In this embodiment, each intervention technique in the psychological expert corpus knowledge graph points to more than two client states, and each client state is pointed to by more than two intervention techniques. Therefore, it is possible to use the current client state as the target, and to retrieve multiple new intervention techniques that point to the target and have a positive effect in the psychological expert corpus knowledge graph. Then, according to the probability weight of the causal chain path of each new intervention technique pointing to the target, the multiple new intervention techniques are sorted from high to low to generate a candidate intervention list.

[0026] S4. Keeping the current symptoms and risk factors unchanged, replace the current intervention with a new intervention technique and simulate and deduce the new probability weights of the new intervention technique pointing to the current state of the visitor in a structured causal model.

[0027] After generating the candidate intervention list, the current symptoms and current risk factors in the structured entity fields are kept unchanged. The current intervention technique is replaced one by one with multiple new intervention techniques from the candidate intervention list. The new probability weights of each new intervention technique pointing to the current visitor's state are simulated and deduced.

[0028] S5. Compare the new probability weights derived in step S4 with the probability weights retrieved in step S2. If the new probability weights derived in step S4 are higher, then the causal chain path of the corresponding new intervention technology pointing to the current state of the client is converted into psychological counseling supervision text output.

[0029] After obtaining the new probability weights of each new intervention technique pointing to the current client's state, the new probability weights derived in step S4 are compared with the probability weights retrieved in step S2. If the new probability weight derived in step S4 is higher, it means that the new intervention technique has a higher probability of effective intervention. Therefore, a large language model is used as a natural language generator to transform the causal chain path of the new intervention technique with the highest probability weight pointing to the current client's state into psychological counseling supervision text output. If the probability weight retrieved in step S2 is higher, it means that the original intervention technique has a higher probability of effective intervention. Therefore, a large language model is used as a natural language generator to transform the corresponding causal chain path of the current intervention technique pointing to the current client's state into psychological counseling supervision text output. The psychological counseling supervision text output in this way combines the causal logic relationship of the change in the client's state caused by the use of intervention techniques, and can provide targeted intervention supervision opinions when facing complex psychological counseling scenarios.

[0030] In this embodiment, the structured causal model is configured to receive structured variables after semantic parsing, and based on, for example... Figure 2The causal ladder theory shown performs the following three levels of operations: L1 Observation Layer: The semantic parsing module analyzes the text and extracts entities, including symptoms, intervention techniques, client status, and risk factors. It retrieves the causal chain path from intervention techniques to client status within the retrieval graph and reads the probability weights along that path.

[0031] L2 Intervention Layer: By querying the visitor's state, the knowledge graph is searched in reverse along the causal chain path to find all new intervention techniques that have a positive effect on the visitor's state. These techniques are then sorted according to the probability weight of the causal chain and used as hypothetical intervention actions.

[0032] L3 Counterfactual Layer: Keeping the symptoms and risk factors unchanged, the new intervention techniques from L2 are substituted into the entity, and the new probability weights obtained by adopting the new actions are recalculated in the knowledge graph. If the new probability weights of the new intervention techniques are higher than the probability weights of L1, then the hypothesis is valid.

[0033] The following is an example of how this embodiment is applied to a high-risk crisis intervention scenario: The system receives psychological counseling dialogues and first extracts keywords from the dialogue text between the counselor and the client using a semantic parser. This is then converted into structured entity fields defined by a knowledge graph. The system also reads the "effective intervention probability weight" of the current technology for the current client's state, stored in the graph. Specifically: the system receives the client's text "Living is a burden," which is mapped to the graph entity symptom "despair" and the risk factor "suicidal tendency." Simultaneously, it receives the counselor's text "I hear you're really suffering," which is mapped to the entity intervention technique "emotional support." The system queries the psychological expert knowledge graph and finds that under the risk factor "suicidal tendency" node, the effectiveness probability weight of the intervention technique "emotional support" pointing to the client's state of "emotional improvement" is marked as 0.15, below the threshold of 0.4.

[0034] Then, a reverse search was performed on the psychological expert corpus knowledge graph, traversing all intervention technique nodes pointing to the client's state of "improved mood," generating a list of hypothetical interventions. Specifically, the system used the client's state of "improved mood" as the target and searched backwards for new intervention techniques with positive effects. The search results showed that the intervention technique "crisis assessment" had a probability weight of 0.72, and the intervention technique "establishing a safety contract" had a probability weight of 0.68, and were listed as candidate interventions.

[0035] Then, keeping the client's current symptoms and risk factors unchanged, the counselor's intervention techniques were replaced with a list of candidate interventions in the virtual computing layer. The probabilities of the consequences that different new intervention techniques might produce in the next moment were calculated. Specifically: Simulating the execution of the new intervention technique "crisis assessment," the results showed that the effective probability weight of this intervention technique pointing to the client's "emotional improvement" was 0.92; Simulating the execution of the new intervention technique "establishing a safety contract," the results showed that the effective probability weight of this intervention technique pointing to the client's "emotional improvement" was 0.78. Compared to maintaining the original intervention technique "emotional support" with a probability weight of 0.15, the probability weight of executing the new intervention technique "crisis assessment" was significantly increased.

[0036] Then, using a natural language generator, the causal chain path corresponding to the new intervention technology "crisis assessment" is transformed into natural language feedback output based on the above judgment results. Specifically: the system outputs the instruction "High-risk signal detected. Current emotional support is ineffective. Please stop comforting immediately and proceed with the suicide crisis assessment process."

[0037] The above are merely embodiments of the present invention and are not intended to limit the scope of patent protection. Any non-substantial changes or substitutions made by those skilled in the art based on the present invention will still fall within the scope of patent protection.

Claims

1. A method for generating psychological counseling case supervision based on causal models and knowledge graphs, characterized in that... This includes the following steps achieved using a structured causal model: S1. Obtain structured entity fields converted from psychological counseling dialogue text, wherein the structured entity fields include at least: symptoms, intervention techniques, client status, and risk factors; S2. Based on the structured entity fields, query the preset psychological expert corpus knowledge graph to obtain the causal chain path and its probability weight of the current intervention technology pointing to the current visitor's state; S3. Using the current state of the visitor as the target, a new intervention technique that points to the target and has a positive effect is retrieved from the knowledge graph of the psychological expert corpus; S4. Keeping the current symptoms and current risk factors unchanged, replace the current intervention technique with the new intervention technique, and simulate and deduce the new probability weights of the new intervention technique pointing to the current state of the visitor in a structured causal model; S5. Compare the new probability weights derived in step S4 with the probability weights retrieved in step S2. If the new probability weights derived in step S4 are higher, then the causal chain path of the corresponding new intervention technology pointing to the current state of the client is converted into psychological counseling supervision text output.

2. The method for generating psychological counseling case supervision according to claim 1, characterized in that, In step S5, if the probability weight obtained in step S2 is higher, the causal chain path of the current intervention technology pointing to the current visitor's state is converted into psychological counseling supervision text output.

3. The method for generating psychological counseling case supervision according to claim 1, characterized in that, In step S3, if multiple new intervention techniques are obtained through reverse retrieval, the multiple new intervention techniques are sorted according to the probability weight of the causal chain path pointing to the target by each new intervention technique to generate a candidate intervention list; in step S4, the current intervention technique is replaced one by one with the multiple new intervention techniques in the candidate intervention list; in step S5, the causal chain path pointing to the current client's state by the new intervention technique with the highest probability weight is converted into psychological counseling supervision text output.

4. The method for generating psychological counseling case supervision according to claim 1, characterized in that, In step S2, the probability weights are obtained by calculating the average causal effect based on the desensitized psychological counseling records. Specifically, natural language processing technology is used to calculate the emotional score of each dialogue in the psychological counseling records as a confounding variable for the quality of the counselor-client relationship. While keeping the confounding variable unchanged, the difference between the probability of the client's emotional improvement when using a specific intervention technique and when not using the specific intervention technique is compared as the average causal effect.

5. The method for generating psychological counseling case supervision according to claim 1, characterized in that, In step S1, a large language model is specifically used as a semantic parser to convert psychological counseling dialogue text into structured entity fields.

6. The method for generating psychological counseling case supervision according to claim 1, characterized in that, In step S5, a large language model is specifically used as a natural language generator to transform the new causal chain path and its probability weights that point the new intervention technology to the current state of the client into psychological counseling supervision text output.

7. A computer program product comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps in the psychological counseling case supervision generation method according to any one of claims 1 to 6.

8. A psychological counseling case supervision generation system based on causal models and knowledge graphs, characterized in that... The system includes a server, which comprises a processor and a memory interconnected thereto, the memory storing the computer program product of claim 7.

Images