User portrait construction method based on knowledge graph and psychological knowledge pushing method

Abstract

According to the knowledge graph-based user portrait construction method and the psychological knowledge pushing method provided by the invention, in a numerous and jumbled knowledge graph, data generated by active behaviors of a user is utilized to construct a dominant user portrait of the user; the hidden user portrait of the user is constructed according to behaviors of other users associated with the user, so that a more comprehensive and more accurate psychological portrait of the user is formed, useful personalized psychological knowledge is accurately pushed to the user, the user is helped to find and solve possible psychological problems as soon as possible, and the user experience is improved. And psychological diseases are avoided.

Description

Technical Field

[0001] The present invention belongs to the technical fields of big data and artificial intelligence, and particularly relates to a method for constructing a user portrait based on a knowledge graph and a method for pushing psychological knowledge. Background Art

[0002] With the development of social economy, on the one hand, people not only pay more and more attention to physical health, but also pay more and more attention to mental health; on the other hand, the pressure of people's study, work and life is increasing day by day, which has a negative impact on the mental health of many people. The development of mental illness often has a gradual process. When people realize that they have mental illness and need treatment, it often takes a long time. On the one hand, it has caused irreversible harm to personal physical and mental health, and on the other hand, it is not easy to be completely cured. Therefore, if people can be actively guided to realize the problems before they have mental illness, and actively solve or at least relieve these problems, it can avoid relatively minor mental problems from further developing into serious mental illness, which is not only of great benefit to personal physical and mental health, but also crucial for maintaining social stability and harmony.

[0003] With the development of big data and artificial intelligence technologies, when people use mobile phones or computers in daily life, mobile applications or computer programs often collect user information and push content that users are interested in according to artificial intelligence algorithms. For example, mobile applications such as Douyin, Kuaishou, and Toutiao will push video or news content according to user preferences; mobile applications such as JD.com, Taobao, and Pinduoduo will also push product links or advertisements according to user preferences or needs. This kind of personalized content push is generally realized based on user portraits. By analyzing a large amount of user data information, abstracting the data into tags, and then using these tags to concretize the user image, a user portrait is finally formed. However, for some people with psychological problems, it is often very difficult to obtain their user portraits because they are "lazy" to actively express their feelings and actively seek solutions. For example, before potential depression patients develop into depression, they often feel that doing anything is meaningless and uninteresting. They are "lazy" to socialize in real life, so they lack chat data with others as the basis for the user portrait; they are also "lazy" to use mobile applications or computer programs on the Internet. Even if they use them, they never actively express their opinions or communicate with others, so they also lack this data as the basis for their user portraits. It can be seen that the existing technology lacks a method for helping these people who already have psychological problems but have not developed into mental illness to construct a comprehensive and accurate user portrait, and this part of the population is precisely the one that most needs attention and help. Moreover, the number of this part of the population is relatively large. If they can be guided in the early stage of psychological problems, the effect of avoiding the development into mental illness is often the best. Summary of the Invention

[0004] In view of the above defects of the prior art, the purpose of the present invention is to provide a method for constructing a user profile based on a knowledge graph and a method for pushing psychological knowledge, which not only uses the data generated by the user's active behavior to construct an explicit user profile in a huge knowledge graph, but also constructs an implicit user profile according to the behaviors of other users associated with the user, so as to form a more comprehensive and accurate user psychological profile, and further realize the accurate push of useful personalized psychological knowledge for the user, so as to help the user discover and relieve existing psychological problems as early as possible and avoid developing into mental diseases. It should be noted that the present invention aims to discover possible psychological problems of users by constructing user psychological profiles and predict the potential risks of their developing into mental diseases, rather than diagnosing or treating the mental diseases of users. If you need to diagnose or treat mental diseases, you still need to seek the help of professionals or professional institutions.

[0005] To achieve the above object, on the one hand, the present invention provides a method for constructing a user psychological profile based on a knowledge graph, including the following steps:

[0006] (1) Collect user information of mobile applications and / or computer programs to form a description text of the user; extract the emotional attributes and attribute values of the user according to the description text to form a description text with emotional information.

[0007] Furthermore, the sources of user information include, but are not limited to: texts in the user registration materials (including real-name information) of mobile applications and / or computer programs, such as basic information like name, online name, ID number, age, mobile phone number, major, job, grade, email, personal signature, personality tags, friend list, mobile phone address book, etc.; content generated by the user during the use of mobile applications and / or computer programs, such as text content like chat records, comments, dynamics, diaries, blogs, etc.; texts that have a connection relationship with the user, such as text content read by the user, video content browsed by the user, etc. It should be noted that the above collection of user information requires the authorization of the user or the official, and the present invention opposes any practice of collecting the above user information through illegal or unauthorized means.

[0008] Furthermore, among the sources of the above user information, non-text information will first be converted into text information, and parameters related to emotions such as speech rate, volume, and pitch will be preserved as much as possible. Then, the emotional attributes and their values of the user will be extracted from the above text information (including parameters related to emotions) to form a descriptive text with emotional information. For example, in a chat record, the user may send a voice message belonging to audio information or a selfie video belonging to video information (which includes audio information). When converting the audio information or video information into text information, parameters related to the user's emotions are identified, such as speech rate (e.g., words per minute, percentage above / below the average speech rate), volume (e.g., decibel value, percentage above / below the average volume), pitch (e.g., high flat tone, mid flat tone, low flat tone, rising tone, falling tone), etc., which can reflect emotions. Converting audio information or video information into text information uses speech recognition technology, which is a very mature existing technology and already has high recognition efficiency and accuracy. Extracting emotional attributes from text information also belongs to existing technology, and most existing general artificial intelligence large models can achieve this, such as ChatGPT of OpenAI, Bing AI of Microsoft, Claude of Anthropic, Grok of xAI, Llama of Meta, Doubao of ByteDance, Wenxin Yiyan of Baidu, Tongyi Qianwen of Alibaba, Pangu Model of Huawei, Hunyuan Model of Tencent, Spark Cognitive Model of iFlytek, Kimi of DarkSide of the Moon, etc.

[0009] Meanwhile, there are also many existing technologies for identifying a user's emotions through audio, such as Chinese patent applications CN109935240A, CN118571265A, etc. If these technologies are used, the audio information or video information (including audio information) does not need to be first converted into text information, but the emotional attributes and their values of the user can be directly extracted from the audio information. Or, there are also many existing technologies for directly identifying a user's emotions through video, such as Chinese patent applications CN108491764A, CN116453024A, CN112492397A, etc. If these technologies are used, the video information does not need to be first converted into audio information or text information, but the emotional attributes and their values of the user can be directly extracted from the video information. Adding these emotional attributes and their values to the user's descriptive text forms a descriptive text with emotional information.

[0010] Furthermore, the above emotional attributes include happiness, anger, sadness, surprise, fear, disgust, jealousy, shyness, guilt, anxiety, depression, etc., and the attribute values represent the degrees of these emotions, and different assignments can be given according to different degrees, such as 0 to 10, and the larger the value, the deeper the degree.

[0011] For example, among the emotions belonging to "sorrow", the emotional intensity of "sadness" is higher than that of "heart - broken", so the attribute assignment of "sadness" is higher than that of "heart - broken". For example, the attribute of the word "heart - broken" is sorrow, and the attribute value is 5; while the attribute of the word "sadness" is also sorrow, but the attribute value is 7.

[0012] Alternatively, the assignment of emotional attributes can also be changed by emphasizing words or weakening words. For example, in Chinese, adverbs such as "very", "super", "extremely", and "… to death" belong to emphasizing words and have the effect of strengthening the attribute assignment; while "a little", "kind of", "slightly" etc. belong to weakening words and have the effect of weakening the attribute assignment. For example, the attribute of the word "heart - broken" is sorrow, and the attribute value is 5; the word "heart - broken to death" has its attribute value adjusted to 6 due to the existence of the emphasizing word "… to death"; while the word "kind of heart - broken" has its attribute value adjusted to 4 due to the existence of the weakening word "kind of".

[0013] Alternatively, parameters such as speech rate, volume, and pitch that can reflect emotions can also change the assignment of emotional attributes. For example, the attribute of the word "happy" is joy, and the attribute value is 5. When the user says a sentence containing "happy" with an increased volume and a higher pitch, the attribute value can be adjusted to 6 or 7 according to the degree of increased volume and higher pitch.

[0014] (2) Extract information from the descriptive text with emotional information to form triples in the form of "entity - relationship - entity" or "entity - attribute - attribute value". The information extraction includes entity recognition, relationship recognition, and attribute recognition, and the information after extraction is stored using a relational database.

[0015] Furthermore, the above - mentioned entity recognition, relationship recognition, and attribute recognition can be implemented by mature artificial intelligence algorithms in the prior art. For example, Chinese Patent Application CN115982379A uses a method combining an improved convolutional neural network and an improved conditional random field model. The descriptive text is input into the improved convolutional neural network to obtain a feature map, and then the feature map is input into the improved conditional random field model for sequence annotation to identify entities of interest in the descriptive text.

[0016] Furthermore, the above - mentioned relational database is selected from one or more of relational databases such as MySQL, Oracle Database, Microsoft SQLServer, PostgreSQL, IBM DB2, MariaDB, SQLite, Informix, etc., and preferably is the MySQL database.

[0017] Furthermore, since the descriptive text with emotional information already contains emotional attributes and their values, the triple of "entity - attribute - attribute value" representing the user's emotion can be directly extracted, that is, the triple of "entity - emotional attribute - attribute value", such as "Zhang San - frustrated - 5"

[0018] Furthermore, for the triple of "entity - attribute - attribute value", the attribute value also contains time information, that is, the generation time of the text information when the triple is extracted. For example, "Li Si - frustrated - 5(202412011505)" indicates that the generation time of this text information is 15:05 on December 1, 2024. Since mental states and mental illnesses develop dynamically, recording the time information corresponding to mental states is very important for judging potential mental illnesses. For example, if a user has potential symptoms of depression during a certain period, then the label of "potential depression" may be attached to their user profile; however, if these symptoms disappear after a period of time, then the label of "potential depression" in their user profile may be deleted or its probability may be lowered. Another example is that bipolar disorder is a mental disorder with both manic and depressive episodes. During the manic episode, patients will show symptoms such as elevated mood, high energy, and increased activity; while during the depressive episode, patients will exhibit symptoms such as low mood, decreased interest, reduced activity, and self - blame. Therefore, potential patients with bipolar disorder may show the above - mentioned opposite symptoms for a relatively long time before developing the disease, which is closely related to time.

[0019] Furthermore, based on the triple of "entity - emotional attribute - attribute value" of the user that contains time information, an emotion map of the user within a certain period (such as 1 minute, 1 hour, 1 day, 1 week, 1 month, 1 year, etc.) can be drawn. In the emotion map, different emotional attributes are marked with different colors, and the larger the attribute value of a certain emotional attribute, the darker the corresponding color. The emotion map can intuitively and visually reflect the user's emotion (the image of the emotion map) or emotion changes (the video formed by the images of multiple emotion maps) within a certain period. Existing artificial intelligence image recognition models can already well recognize and compare these images and videos, such as the deep convolutional neural network with the Residual Network (ResNet) architecture, such as the ResNet - 50 network structure.

[0020] (3) Perform knowledge fusion on the triple, perform co - reference resolution on different triples of the same entity from multiple sources to map to the correct single entity, and disambiguate the triples with the same name representing different entities to solve the ambiguity generated by the triples with the same name.

[0021] Furthermore, the above-mentioned knowledge fusion can be achieved by mature artificial intelligence algorithms in the prior art. For example, an unsupervised clustering method based on encyclopedic knowledge can be used to solve the ambiguity caused by homonymous triples.

[0022] (4) Perform knowledge processing on the fused triples to form a structured and networked knowledge system; use a graph database to store the triples after knowledge processing to form a primary knowledge graph.

[0023] Furthermore, the knowledge processing includes ontology construction, knowledge reasoning, and quality assessment. Among them, the ontology construction extracts a general term based on the common understanding of a field; the knowledge reasoning obtains new knowledge or conclusions through various methods; the quality assessment quantifies the credibility of knowledge and ensures the quality of the knowledge base by discarding knowledge with low confidence.

[0024] Furthermore, the above-mentioned graph database is selected from one or several of the graph databases such as Neo4j, JanusGraph, OrientDB, ArangoDB, Titan, Virtuoso, Stardog, TigerGraph, AllegroGraph, Amazon Neptune, HugeGraph, GeaBase, etc., and preferably the Neo4j database.

[0025] (5) On the basis of the primary knowledge graph, based on a certain person entity (i.e., the user), extract the triples related to the mental state generated by its active behavior to form an explicit knowledge graph about the person entity; extract the triples related to the mental state of the person entity from the triples of other person entities associated with it to form an implicit knowledge graph about the person entity.

[0026] Furthermore, the triples related to the mental state generated by the above-mentioned active behavior refer to the triples of keywords related to the mental state actively expressed, described, or evaluated by the person entity itself.

[0027] Furthermore, the triples related to the mental state of the person entity in the triples of other person entities associated with it refer to the triples of keywords related to the mental state of the person entity expressed, described, or evaluated by others (the others have a more or less direct or indirect relationship with the person entity).

[0028] Furthermore, the above-mentioned explicit knowledge graph is like the trunk and branches of a big tree, which is above the ground and represents the subjective expression, description or evaluation of a person entity's own mental state; while the implicit knowledge graph is like the root part of this big tree, which is below the ground and represents the objective expression, description or evaluation of the mental state of the people around this person entity (who have a direct or indirect relationship with this person entity, whether far or near). Since everyone's self-awareness is different in reality, a person's self-subjective evaluation is often one-sided or even wrong, and often different from the evaluation made by others. Also, due to the complexity and multi-faceted nature of human nature, a person often shows completely different states in front of different groups of people. Therefore, the evaluations of a person by others with different degrees of closeness and distance are often different or even contradictory. However, the existing psychological knowledge graph does not distinguish between subjective and objective mental state evaluations, so the user portraits constructed based on this are often not comprehensive and accurate enough.

[0029] (6) Mine explicit user characteristics based on the explicit knowledge graph, and construct an explicit user portrait of this person entity according to the explicit user characteristics; mine implicit user characteristics based on the implicit knowledge graph, and construct an implicit user portrait of this person entity according to the implicit user characteristics; the explicit user portrait and the implicit user portrait together constitute the user psychological portrait of this person entity, and the user psychological portrait includes user psychological tags.

[0030] Furthermore, regarding how to mine user characteristics and construct user portraits through knowledge graphs, there are already mature artificial intelligence algorithms in the existing technology to achieve this, such as Chinese Patent Application CN115982379A.

[0031] Furthermore, each user psychological tag contains a corresponding probability value, indicating the probability that the user may potentially suffer from the corresponding mental illness. Please note that the purpose of this invention is to discover potential psychological problems by constructing a complete user psychological portrait and predict and prevent the potential risk of developing into a mental illness, rather than diagnosing the user's mental illness. If you need to diagnose or treat mental illness, you still need to seek help from professionals or professional institutions.

[0032] Furthermore, the above-mentioned user psychological tags (including probability values) can be obtained by an artificial intelligence image recognition model to recognize and compare the user's emotion map. For example, the present invention also provides a method for using a deep convolutional neural network with a deep residual network architecture to recognize and compare the user's emotion map to obtain user psychological tags, including the following steps:

[0033] A. Take users diagnosed with various psychological diseases (diagnosed by professionals or professional institutions) as sample users, and obtain the emotional maps (including time information) of the sample users within a period of time before and after diagnosis (for example, 2 years before illness and 1 year after diagnosis) as the sample data set;

[0034] B. Establish the backbone network of the artificial intelligence data analysis module, and use the sample data set to pre-train the backbone network of the artificial intelligence data analysis module;

[0035] C. Transfer the weights of the pre-trained backbone network of the artificial intelligence data analysis module to the analysis network of the artificial intelligence data analysis module, and iteratively train the analysis network on the sample data set;

[0036] D. Obtain the emotional map of the user to be analyzed within a period of time, and use the emotional map (including time information) of the user to be analyzed as the input of the analysis network obtained by iterative training in step D to obtain the psychological label of the user to be analyzed, that is, the probability that the user to be analyzed may potentially suffer from various psychological diseases. In essence, it is the similarity between the psychological map of the user to be analyzed and the psychological map of the sample user.

[0037] On the other hand, the present invention also provides a method for pushing psychological knowledge based on the above user psychological portraits (including explicit user portraits and implicit user portraits), including the following steps:

[0038] S1: Extract the explicit psychological labels and implicit psychological labels related to the psychological state from the labels of the explicit user portrait and the implicit user portrait of the same user respectively, and compare whether there are contradictions between the explicit psychological labels and the implicit psychological labels: If there are no contradictions, push the psychological knowledge related to the explicit psychological label with the highest probability and the implicit psychological label with the highest probability; If there are contradictions, enter S2;

[0039] Furthermore, whether there are contradictions between the above psychological labels refers to whether the psychological diseases corresponding to the psychological labels have completely opposite and incompatible symptoms. If so, it means that the two psychological diseases are generally contradictory and incompatible, and the user may suffer from schizophrenia. For example, patients with histrionic personality disorder tend to be overly emotional and dramatic in social situations; while patients with schizoid personality disorder are extremely cold and aloof in social aspects, and their symptoms are opposite and incompatible. If the explicit psychological label of a certain user is "histrionic personality disorder", but its implicit psychological label is "schizoid personality disorder", then the user may suffer from schizophrenia.

[0040] S2: Push psychological knowledge about schizophrenia to this user and other users associated with this user, and compare the probabilities of the psychological tags in the explicit user portrait and the implicit user portrait of this user. If the probability of the explicit psychological tag with the highest probability in the explicit user portrait is much greater than the probability of the implicit psychological tag with the highest probability in the implicit user portrait, then push the psychological knowledge related to the explicit psychological tag with the highest probability to this user; if the probability of the implicit psychological tag with the highest probability in the implicit user portrait is much greater than the probability of the explicit psychological tag with the highest probability in the explicit user portrait, then push the psychological knowledge related to the implicit psychological tag with the highest probability to this user; if the probability of the explicit psychological tag with the highest probability in the explicit user portrait is equivalent to or the same as the probability of the implicit psychological tag with the highest probability in the implicit user portrait, then push both the psychological knowledge related to the explicit psychological tag with the highest probability and the psychological knowledge related to the implicit psychological tag with the highest probability to this user.

[0041] Further, the above "much greater than" means that the probability (larger) of the psychological tag in one user portrait is 2 times or more of the probability (smaller) of the psychological tag in the other user portrait. The above "equivalent probability" means that the probability (larger) of the psychological tag in one user portrait is 1 - 2 times (excluding 2 times) of the probability (smaller) of the psychological tag in the other user portrait.

[0042] S3: If the number and / or probability of the explicit psychological tags and the implicit psychological tags of this user are both 0, then push the psychological knowledge about psychological diseases such as depression and bipolar disorder (especially during the depressive episode) to this user and other users associated with this user.

[0043] The user portrait construction method and psychological knowledge push method based on the knowledge graph of the present invention have at least the following beneficial technical effects:

[0044] (1) The present invention innovatively collects the user information of mobile applications and / or computer programs to form a description text of the user, extracts the emotional attributes and attribute values of the user from it to form a description text with emotional information, and then draws an emotional map of the user; and uses a deep convolutional neural network with a residual network architecture to identify and compare the emotional map of the user, so as to obtain user psychological tags with probability values, that is, the probabilities of the user potentially suffering from various psychological diseases, realizing the quantitative prediction of the user's potential disease probability.

[0045] (2) Based on the complex primary knowledge graph, not only the data generated by the user's active behavior is used to construct the explicit user portrait of the user, but also the behavior of other users associated with the user is used to construct the implicit user portrait of the user, so as to form a more comprehensive and accurate user psychological portrait, and then realize the precise push of useful personalized psychological knowledge for the user to help prevent and solve the user's psychological problems and avoid developing into mental diseases.

[0046] (3) Extract the explicit psychological labels and implicit psychological labels from the labels of the explicit user portrait and the implicit user portrait of the same user respectively, and judge whether there are contradictions and which psychological label dominates, so as to realize the precise push of the most useful personalized psychological knowledge for the user.

[0047] (4) For the "invisible" users whose numbers and / or probabilities of both explicit psychological labels and implicit psychological labels are 0, corresponding psychological knowledge push schemes are also provided to help them to the greatest extent. Description of the Drawings

[0048] Figure 1 is an example of a chat record generated by a user during the use of a mobile application in a preferred embodiment of the present invention;

[0049] Figure 2 is the emotional map of a user within 7 hours in a preferred embodiment of the present invention;

[0050] Figure 3 is the emotional map of a user within 7 days in a preferred embodiment of the present invention;

[0051] Figure 4 is an example of the primary knowledge graph formed in a preferred embodiment of the present invention;

[0052] Figure 5 is a schematic diagram of the explicit knowledge graph and the implicit knowledge graph formed in a preferred embodiment of the present invention;

[0053] Figure 6 is a schematic flowchart of the method for constructing a user psychological portrait based on a knowledge graph in a preferred embodiment of the present invention;

[0054] Figure 7 is a schematic flowchart of the method for pushing psychological knowledge based on a user portrait in a preferred embodiment of the present invention. Detailed Embodiments

[0055] The following details the embodiments of the present invention. The following embodiments are implemented on the premise of the technical solution of the present invention, and detailed implementation manners and specific operation processes are given. However, the protection scope of the present invention is not limited to the following embodiments.

[0056] As Figure 4 shown, in a preferred embodiment, the present invention provides a method for constructing a user psychological portrait based on a knowledge graph, comprising the following steps:

[0057] (1) Collect user information of mobile applications and / or computer programs to form a description text of the user; extract the emotional attributes and their values of the user from the description text to form a description text with emotional information.

[0058] The sources of user information include, but are not limited to: texts in the user registration materials (including real-name information) of mobile applications and / or computer programs, such as basic information like name, online name, ID number, mobile phone number, major, job, age, grade, email, personal signature, personal label, friend list, mobile phone address book, etc.; content generated by the user during the use of mobile applications and / or computer programs, such as text content like chat records, published comments, dynamics, diaries, blogs, etc.; texts that have a connection relationship with the user, such as text content read by the user, video content browsed by the user, etc.

[0059] Among the above sources of user information, non-text information will first be converted into text information, and parameters related to emotions such as speech rate, volume, and pitch will be retained as much as possible. Then, the emotional attributes and their values of the user will be extracted from the above text information (including parameters related to emotions) to form a description text with emotional information. For example, in a chat record, the user may send a voice message belonging to audio information or a selfie video belonging to video information (which includes audio information). When converting the audio information or video information into text information, the emotions of the user are identified or parameters that can reflect emotions such as speech rate (e.g., words per minute, percentage higher / lower than the average speech rate), volume (e.g., decibel value, percentage higher / lower than the average volume), and pitch (e.g., high flat tone, medium flat tone, low flat tone, rising tone, falling tone) are retained. Converting audio information or video information into text information uses speech recognition technology, which is a very mature existing technology and already has a high recognition efficiency and accuracy. Extracting emotional attributes from text information also belongs to the existing technology, and most existing general artificial intelligence large models can achieve it, such as ChatGPT of OpenAI, Bing AI of Microsoft, Claude of Anthropic, Grok of xAI, Llama of Meta, Doubao of ByteDance, Wenxin Yiyan of Baidu, Tongyi Qianwen of Alibaba, Pangu Model of Huawei, Hunyuan Model of Tencent, Spark Cognitive Model of iFlytek, Kimi of MoonDarkSide, etc.

[0060] The above-mentioned emotional attributes include happiness, anger, sadness, surprise, fear, disgust, jealousy, shyness, guilt, anxiety, depression, etc., and the attribute values represent the degrees of these emotions. Different assignments can be given according to different degrees. For example, from 0 to 10, the larger the value, the deeper the degree.

[0061] For example, among the emotions belonging to "sadness", the emotional intensity of "sorrow" is higher than that of "heartbreak". Therefore, the attribute assignment of "sorrow" is higher than that of "heartbreak". For example, the attribute of the word "heartbreak" is sadness, and the attribute value is 5; while the attribute of the word "sorrow" is also sadness, but the attribute value is 7.

[0062] Or, the assignment of emotional attributes can also be changed by emphasizing words or weakening words. For example, in Chinese, adverbs such as "very", "super", "extremely", "... to death" belong to emphasizing words and have the effect of strengthening the attribute assignment; while "a little", "kind of", "slightly" etc. belong to weakening words and have the effect of weakening the attribute assignment. For example, the attribute of the word "heartbreak" is sadness, and the attribute value is 5; the word "heartbroken to death" has an attribute value adjusted to 6 due to the emphasizing word "... to death"; while the word "kind of heartbroken" has an attribute value adjusted to 4 due to the weakening word "kind of".

[0063] Or, parameters such as speech rate, volume, and pitch that can reflect emotions can also change the assignment of emotional attributes. For example, the attribute of the word "happy" is happiness, and the attribute value is 5. When the user says a sentence containing "happy" with a louder volume and a higher pitch, the attribute value can be adjusted to 6 or 7 according to the degree of increased volume and increased pitch.

[0064] In a preferred embodiment, as Figure 1 shown, user Li Si generated a chat record during the use of the mobile application, which contains both text information and voice information. The voice information will first be converted into text information (or with emotion-related parameters) through speech recognition for storage. Then, identify the emotional attributes carried in the text information and assign values to these emotional attributes. For example, in Li Si's words "Mom, I'm so annoyed recently", "so annoyed" can be given emotional attributes of anxiety, depression, and anger, and the attribute values are 5, 3, and 1 respectively. Then the above text information can be expressed as "Mom, I'm so annoyed recently (anxiety 5, depression 3, anger 1)". Emotions are usually a direct manifestation of the user's mental state and can be used to assist in judging whether the user has mental problems. Therefore, a large amount of user data related to emotions can assist in judging the possibility of the user potentially suffering from various mental diseases.

[0065] (2) Extract information from the described text with emotional information to form triples in the form of "entity-relationship-entity" or "entity-attribute-attribute value". The information extraction includes entity recognition, relationship recognition, and attribute recognition, and the information after information extraction is stored using a relational database.

[0066] The above entity recognition, relationship recognition, and attribute recognition can be implemented by mature artificial intelligence algorithms in the prior art. For example, Chinese Patent Application CN115982379A uses a method of combining an improved convolutional neural network with an improved conditional random field model. The described text is input into the improved convolutional neural network to obtain a feature map, and the feature map is then input into the improved conditional random field model for sequence labeling to identify entities of interest in the described text.

[0067] The above relational database is selected from one or several of relational databases such as MySQL, Oracle Database, Microsoft SQL Server, PostgreSQL, IBM DB2, MariaDB, SQLite, Informix, etc., and preferably is a MySQL database.

[0068] For example, for the above Figure 1 chat record, information extraction can be performed to extract multiple "entity-relationship-entity" triples such as "Zhang San - mother-son - Li Si", "Li Si - lover - Wang Wu", "Zhang San - neighbor - Zhao Liu", etc.; or multiple "entity-attribute-attribute value" triples such as "Zhang San - surprised - 1", "Li Si - anxious - 5", "Li Si - depressed - 3", "Li Si - angry - 1", etc.

[0069] For the "entity-attribute-attribute value" triple, the attribute value also includes time information. For example, the attribute value of the above "Zhang San - surprised - 1" triple also includes time information, that is, the generation time of the text information when this triple is extracted, which is also the sending time of the sentence "Si Zai, why haven't you made video calls with your parents recently?". If the sending time is 18:06 on September 28, 2024, then the above "Zhang San - surprised - 1" triple can be saved as "Zhang San - surprised - 1(202409281806)" after including the time information.

[0070] Since mental states and mental illnesses are dynamically evolving, recording the time information corresponding to mental states is very important for judging potential mental illnesses. For example, if a user has potential symptoms of depression during a certain period, then the label of "potential depression" may be attached to their user profile; however, after a period of time, these symptoms disappear, then the label of "potential depression" in their user profile may be deleted or its probability may be lowered. Another example is that bipolar disorder is a mental disorder with both manic and depressive episodes. During the manic episode, patients will show symptoms such as elevated mood, high energy, and increased activity; while during the depressive episode, patients will experience symptoms such as low mood, loss of interest, decreased activity, and self-blame, so potential patients with bipolar disorder may show the above-mentioned diametrically opposite symptoms for a relatively long time before developing the disease, which is closely related to time.

[0071] Based on the "entity - emotion attribute - attribute value" triple containing time information of the user, an emotion map of the user for a certain period (such as 1 minute, 1 hour, 1 day, 1 week, 1 month, 1 year, etc.) can be drawn. In the emotion map, different emotion attributes are marked with different colors, and the greater the attribute value of a certain emotion attribute, the deeper the corresponding color. The emotion map can intuitively and visually reflect the user's emotions (the image of the emotion map) or emotional changes (the video formed by the images of multiple emotion maps) within a certain period. Existing artificial intelligence image recognition models can already well recognize and compare these images and videos, such as the deep convolutional neural network with a deep residual network architecture, such as the ResNet - 50 network structure. For example, Figure 2 and Figure 3 are the emotion maps of a certain user within 7 hours (12:00 - 18:00 on December 1, 2024) and 7 days (December 1 - December 7, 2024) respectively. The emotion maps at multiple smaller time scales can be added up to obtain an emotion map at a larger time scale.

[0072] (3) Perform knowledge fusion on the triple, perform coreference resolution on different triples about the same entity from multiple sources to map to the correct single entity, and disambiguate the homonymous triples representing different entities to solve the ambiguity generated by homonymous triples.

[0073] The above-mentioned knowledge fusion can be achieved by mature artificial intelligence algorithms in the existing technology. For example, an unsupervised clustering method based on encyclopedic knowledge can be used to solve the ambiguity generated by homonymous triples.

[0074] For example, by performing coreference resolution on the triples extracted from the chat record, it can be recognized that "Sizai" and "Li Si" co-refer to the same entity, so they can be mapped to the same entity Li Si; "Wang Wu" and "Xiao Wang" co-refer to the same entity, so they can be mapped to the same entity Wang Wu.

[0075] (4) Perform knowledge processing on the fused triples to form a structured and networked knowledge system; use a graph database to store the triples after knowledge processing to form a primary knowledge graph.

[0076] The knowledge processing includes ontology construction, knowledge reasoning, and quality assessment. Among them, the ontology construction extracts a general term based on the common understanding of a field; the knowledge reasoning obtains new knowledge or conclusions through various methods; the quality assessment quantifies the credibility of the knowledge and ensures the quality of the knowledge base by discarding knowledge with low confidence.

[0077] The above graph database is selected from one or several of the graph databases such as Neo4j, JanusGraph, OrientDB, ArangoDB, Titan, Virtuoso, Stardog, TigerGraph, AllegroGraph, Amazon Neptune, HugeGraph, GeaBase, etc., preferably the Neo4j database.

[0078] Figure 4 is part of the primary knowledge graph formed by Figure 1 the chat record. Among them, the "insomnia", "difficulty getting up", "poor appetite", "exercise inertia", etc. shown by Li Si belong to health states or physical symptoms, and these physical symptoms may be related to physical or mental diseases and can be used to assist in judging whether the user may have psychological problems or a potential tendency to develop mental diseases. What Li Si said, "unlucky", is a self-evaluation of his own state, which can also reflect the user's mental state, so it can also be used to assist in judging whether the user has psychological problems or a potential tendency to develop mental diseases. The "depression tendency" shown by Zhao Liu is an evaluation of others made in the conversation between Zhang San and Li Si, and can also be used to assist in judging whether Zhao Liu has psychological problems or a potential tendency to develop mental diseases.

[0079] (5) Based on the primary knowledge graph, for a certain person entity (i.e., the user), extract the triples related to the mental state generated by his active behavior to form an explicit knowledge graph about the person entity; extract the triples related to the mental state of the person entity from the triples of other person entities associated with it to form an implicit knowledge graph about the person entity.

[0080] The triple related to the mental state generated by the above-mentioned active behavior refers to the triple of keywords related to the mental state that are actively expressed, described, or evaluated by the person entity itself. For example, the sentence "Mom, I'm so annoyed recently" is Li Si's own active emotional expression, and "so annoyed" is a keyword related to the mental state. Another example, "Why am I so unlucky recently" is a self-evaluation of one's own state, which can also reflect the user's mental state, and "unlucky" is a keyword related to the mental state.

[0081] Among the triples of other person entities associated with it, the triple related to the mental state of this person entity refers to the triple of keywords related to the mental state of this person entity that are expressed, described, or evaluated by others (the others have a more or less direct or indirect relationship with this person entity). For example, the sentence "Sizi, mom knows you're under a lot of pressure" is said by Zhang San directly to Li Si, and "under a lot of pressure" is Zhang San's evaluation of Li Si, which is a keyword related to the mental state. Another example, the sentence "Almost depressed" is an evaluation of Zhao Liu made by Zhang San in the conversation with Li Si, and "depressed" is an evaluation of Zhao Liu made by Zhang San, which is a keyword related to the mental state.

[0082] Taking Figure 5 the big tree shown as an example, the above-mentioned explicit knowledge graph 1 is like the trunk and branches of the big tree, which is above the ground and represents the subjective expression, description, or evaluation of a person entity's mental state by oneself; while the implicit knowledge graph 2 is like the root of the big tree, which is below the ground and represents the objective expression, description, or evaluation of the mental state of the people around this person entity (who have a more or less direct or indirect relationship with this person entity). Due to the different self-cognitions of each person in reality, a person's self-subjective evaluation is often one-sided or even wrong, and often different from the evaluation made by others. Also due to the complexity and multi-faceted nature of human nature, a person often shows completely different states in front of different groups of people. Therefore, the evaluations of a person by others with different degrees of intimacy are often different or even contradictory. However, the existing psychological knowledge graph in the prior art does not distinguish between subjective and objective mental state evaluations, so the user portraits constructed based on this are often not comprehensive and accurate enough.

[0083] (6) Mine explicit user characteristics based on the explicit knowledge graph, and construct an explicit user portrait of this person entity according to the explicit user characteristics; mine implicit user characteristics based on the implicit knowledge graph, and construct an implicit user portrait of this person entity according to the implicit user characteristics, and the user psychological portrait includes user psychological labels.

[0084] Regarding how to mine user characteristics and construct user portraits through knowledge graphs, there are already mature artificial intelligence algorithms in the prior art to achieve this, such as Chinese Patent Application CN115982379A.

[0085] For example, based on keywords, emotion attributes, and attribute values related to psychological states such as "insomnia", "difficulty getting up", "poor appetite", "exercise inertia", "depression", "bad luck", and "tendency to break up" in Li Si's explicit knowledge graph, the artificial intelligence algorithm can label Li Si's explicit user portrait with explicit psychological labels such as "anxiety disorder", "dysthymia disorder", and "depression". Another example is that based on keywords related to psychological states such as "depression tendency" in Zhao Liu's implicit knowledge graph, the artificial intelligence algorithm can label Zhao Liu's implicit user portrait with implicit psychological labels such as "depression". These psychological labels are the results of qualitative analysis of users based on the above keywords in the knowledge graph.

[0086] In addition to qualitative analysis, each user psychological label of the present invention also includes a corresponding probability value, that is, quantitative analysis, indicating the probability that the user may potentially suffer from the corresponding mental illness. The user psychological label containing the probability value can be obtained by an artificial intelligence image recognition model for recognizing and comparing the user's emotion map. For example, using a deep convolutional neural network with a deep residual network architecture to recognize and compare the user's emotion map, so as to obtain a user psychological label containing a probability value, including the following steps:

[0087] A. Taking users diagnosed with various mental illnesses (diagnosed by professionals or professional institutions) as sample users, obtaining the emotion maps (including time information) of the sample users within a certain period before and after diagnosis (such as 2 years before illness and 1 year after diagnosis) as the sample data set;

[0088] B. Establishing the backbone network of the artificial intelligence data analysis module and pre-training the backbone network of the artificial intelligence data analysis module using the sample data set;

[0089] C. Transferring the weights of the pre-trained backbone network of the artificial intelligence data analysis module to the analysis network of the artificial intelligence data analysis module and iteratively training the analysis network on the sample data set;

[0090] D. Obtain the emotional map of the user to be analyzed over a period of time, and use the emotional map (including time information) of the user to be analyzed as the input of the analysis network obtained by iterative training in step D to obtain the psychological label of the user to be analyzed, that is, the probability that the user to be analyzed may suffer from various mental illnesses. In essence, it is the similarity between the psychological map of the user to be analyzed and the psychological map of the sample user. For example, use the emotional map of Li Si over a period of time as the input of the analysis network obtained by iterative training to obtain Li Si's psychological labels as "anxiety disorder (31%)", "dysthymic disorder (17%)", "depression (8%)", indicating that the probabilities of Li Si potentially suffering from anxiety disorder, dysthymic disorder, and depression are 31%, 17%, and 8% respectively.

[0091] As Figure 7 shown, in a preferred embodiment, the present invention also provides a method for pushing psychological knowledge based on the above user psychological portrait (including the explicit user portrait and the implicit user portrait), including the following steps:

[0092] S1: Extract the explicit psychological labels and implicit psychological labels related to the mental state from the labels of the explicit user portrait and the implicit user portrait of the same user respectively, and compare whether there are contradictions between the explicit psychological labels and the implicit psychological labels: If there are no contradictions, push the psychological knowledge related to the explicit psychological label with the highest probability and the implicit psychological label with the highest probability; If there are contradictions, enter S2;

[0093] Whether there are contradictions between the above psychological labels means whether the mental illnesses corresponding to the psychological labels have completely opposite and incompatible symptoms. If so, it means that the two mental illnesses are generally contradictory and incompatible, and the user may suffer from schizophrenia. For example, patients with histrionic personality disorder tend to be overly emotional and dramatic in social situations; while patients with schizoid personality disorder are extremely cold and alienated in social aspects, and their symptoms are opposite and incompatible. If the explicit psychological label of a certain user is "histrionic personality disorder", but its implicit psychological label is "schizoid personality disorder", then the user may have a potential risk of suffering from schizophrenia.

[0094] S2: Push psychological knowledge about schizophrenia to this user and other users associated with this user, and compare the probabilities of the psychological tags in the explicit user profile and the implicit user profile of this user. If the probability of the most probable explicit psychological tag in the explicit user profile is much greater than the probability of the most probable implicit psychological tag in the implicit user profile, then push the psychological knowledge related to the most probable explicit psychological tag to this user; if the probability of the most probable implicit psychological tag in the implicit user profile is much greater than the probability of the most probable explicit psychological tag in the explicit user profile, then push the psychological knowledge related to the most probable implicit psychological tag to this user; if the probability of the most probable explicit psychological tag in the explicit user profile is equivalent to or the same as the probability of the most probable implicit psychological tag in the implicit user profile, then push both the psychological knowledge related to the most probable explicit psychological tag and the psychological knowledge related to the most probable implicit psychological tag to this user.

[0095] The above "much greater than" means that the probability (larger) of a psychological tag in one user profile is 2 times or more of the probability (smaller) of a psychological tag in another user profile. The above "equivalent probability" means that the probability (larger) of a psychological tag in one user profile is 1 - 2 times (excluding 2 times) of the probability (smaller) of a psychological tag in another user profile.

[0096] For example, if the most probable psychological tag in the explicit user profile of a certain user is "anxiety disorder (30%)", and the most probable psychological tag in its implicit user profile is "dysthymic disorder (10%)", then push the psychological knowledge related to "anxiety disorder" to this user.

[0097] S3: If the number and / or probability of the most probable explicit psychological tag and the most probable implicit psychological tag of a certain user are both 0, then push the psychological knowledge about psychological diseases such as depression and bipolar disorder (especially during the depressive episode) to this user and other users associated with this user.

[0098] The preferred specific embodiments of the present invention have been described in detail above. It should be understood that those of ordinary skill in the art can make many modifications and variations based on the concept of the present invention without creative labor. Therefore, all technical solutions that can be obtained by those skilled in the art in the technical field according to the concept of the present invention through logical analysis, reasoning, or limited experiments on the basis of the prior art should fall within the protection scope determined by the claims.

Claims

1. A method for constructing a user psychological portrait based on a knowledge graph, characterized in that: The following steps are involved: (1) collecting user information of mobile phone applications and / or computer programs to form a description text of the user; extracting the user's emotional attributes and attribute values ​​based on the description text to form a description text with emotional information; (2) extracting information from the description text containing the emotional information to form a triple in the form of "entity-relationship-entity" or "entity-attribute-attribute value", wherein the information extraction includes entity recognition, relationship recognition and attribute recognition, and a relational database is used to store the extracted information; (3) performing knowledge fusion on the triples, performing co-reference resolution on different triples about the same entity from multiple sources to map them to a correct entity, and performing disambiguation on triples with the same name representing different entities to resolve the ambiguity caused by triples with the same name; (4) Perform knowledge processing on the fused triples to form a structured and networked knowledge system; use a graph database to store the knowledge-processed triples to form a primary knowledge graph; (5) Based on the primary knowledge graph, based on a certain character entity, extracting triples related to the psychological state generated by its active behavior to form an explicit knowledge graph about the character entity; extracting triples related to the psychological state of the character entity from triples of other character entities associated with it to form an implicit knowledge graph about the character entity; (6) mining explicit user features based on the explicit knowledge graph, and constructing an explicit user profile of the person entity according to the explicit user features; Based on the implicit knowledge graph, implicit user features are mined, and an implicit user portrait about the character entity is constructed according to the implicit user features; the explicit user portrait and the implicit user portrait together constitute a user psychological portrait about the character entity, and the user psychological portrait includes a user psychological label.

2. The method for constructing a user psychological portrait based on a knowledge graph according to claim 1, characterized in that: The sources of the user information in step (1) include: text in the user registration information of the mobile application and / or computer program; content generated by the user in the process of using the mobile application and / or computer program; text that has a connection with the user; among the sources of the user information, non-text information will first be converted into text information, and parameters related to emotions will be retained; then, the user's emotional attributes and attribute values ​​will be extracted based on the description text to form the description text with emotional information.

3. The method for constructing a user psychological portrait based on a knowledge graph according to claim 2, characterized in that: The user's emotional attributes and attribute values ​​are extracted according to the description text, and the extraction is achieved through a general artificial intelligence big model; the general artificial intelligence big model is selected from one or more of ChatGPT, Bing AI, Claude, Grok, Llama, Doubao, Wenxin Yiyan, Tongyi Qianwen, Pangu Big Model, Hunyuan Big Model, Spark Cognitive Big Model, and Kimi.

4. The method for constructing a user psychological portrait based on a knowledge graph according to claim 1, characterized in that: The relational database in step (2) is selected from one or more of MySQL, Oracle Database, Microsoft SQL Server, PostgreSQL, IBM DB2, MariaDB, SQLite, and Informix relational databases.

5. The method for constructing a user psychological portrait based on a knowledge graph according to claim 1, characterized in that: The attribute value of the triple of "entity-attribute-attribute value" in step (2) also includes time information, that is, the generation time of the text information extracted from the triple.

6. The method for constructing a user psychological portrait based on a knowledge graph according to claim 1, characterized in that: The graph database in step (4) is selected from one or more of Neo4j, JanusGraph, OrientDB, ArangoDB, Titan, Virtuoso, Stardog, TigerGraph, AllegroGraph, Amazon Neptune, HugeGraph, and GeaBase graph databases.

7. The method for constructing user psychological portrait based on knowledge graph according to claim 1, characterized in that: The triples related to the psychological state generated by the active behavior in step (5) refer to the triples of keywords related to the psychological state that are actively expressed, described or evaluated by the character entity itself; the triples related to the psychological state of the character entity in the triples of other character entities associated with it refer to the triples of keywords related to the psychological state that are expressed, described or evaluated by other people who have a relationship with the character entity.

8. The method for constructing a user psychological portrait based on a knowledge graph according to claim 1, characterized in that: The user psychological label in step (6) includes a corresponding probability value, which indicates the probability that the user may potentially suffer from the corresponding psychological disease.

9. A method for pushing psychological knowledge based on the user psychological portrait according to any one of claims 1 to 8, characterized in that: The following steps are involved: S1: Extract explicit psychological labels and implicit psychological labels related to psychological states from the labels of the explicit user portrait and implicit user portrait of the same user, and compare whether there is any contradiction between the explicit psychological labels and implicit psychological labels: if there is no contradiction, push the psychological knowledge related to the explicit psychological label with the highest probability and the implicit psychological label with the highest probability; if there is a contradiction, enter S2; S2: Pushing psychological knowledge about schizophrenia to the user and other users associated with the user, and comparing the probabilities of psychological labels in the explicit user portrait and implicit user portrait of the user. If the probability of the explicit psychological label with the highest probability in the explicit user portrait is much greater than the probability of the implicit psychological label with the highest probability in the implicit user portrait, then pushing psychological knowledge related to the explicit psychological label with the highest probability to the user; if the probability of the implicit psychological label with the highest probability in the implicit user portrait is much greater than the probability of the explicit psychological label with the highest probability in the explicit user portrait, then pushing psychological knowledge related to the implicit psychological label with the highest probability to the user; if the probability of the explicit psychological label with the highest probability in the explicit user portrait is equivalent to or the same as the probability of the implicit label with the highest probability in the implicit user portrait, then pushing both psychological knowledge related to the explicit psychological label with the highest probability and psychological knowledge related to the implicit psychological label with the highest probability to the user; S3: If the number and / or probability of the explicit psychological tags and implicit psychological tags of the user are both 0, then the psychological knowledge about depression and bipolar disorder is pushed to the user and other users associated with the user.

10. The method for pushing psychological knowledge of user psychological portrait according to claim 9, characterized in that: The "much greater than" in step S2 means that the probability (larger) of a psychological label in a certain user portrait is 2 times or more than the probability (smaller) of a psychological label in another user portrait; the "equivalent probability" means that the probability (larger) of a psychological label in a certain user portrait is 1-2 times (not including 2 times) of the probability (smaller) of a psychological label in another user portrait.

Images