Financial institution internal process control method and system based on artificial intelligence

Abstract

The invention relates to the technical field of emotion recognition, and discloses a financial institution internal process control method and system based on artificial intelligence, and the method comprises the following steps: S101, collecting behavior data and character input data when a user uses financial software; step S102, preprocessing the behavior data and the character input data to respectively obtain a first feature sequence and a second feature sequence; s103, constructing an emotion recognition model according to the first feature sequence and the second feature sequence to output an emotion state probability value of the user; and S104, generating a response measure according to the emotional state probability value of the user output by the emotion recognition model. According to the method, the emotional state of the user can be dynamically perceived, financial process adjustment is carried out according to the emotional state, and the personalized requirements of the user in different situations are met.

Description

Technical Field

[0001] The present invention belongs to the technical field of emotion recognition, and in particular relates to an internal process control method and system of a financial institution based on artificial intelligence. Background Art

[0002] In traditional financial service processes, users' operations usually follow pre-set fixed steps. This standardized process design aims to ensure that all users can complete operations in a unified path when conducting financial transactions, purchasing products or inquiring about information, thereby improving the efficiency and consistency of operations. However, although this "one-size-fits-all" process helps to simplify system design, it often ignores the different emotional states that users may have during actual use, such as anxiety, hesitation or confusion. Especially when it comes to complex decisions or high-risk operations, users' emotional fluctuations will directly affect their operating experience and final decision-making behavior. For example, when faced with more complex financial product choices, users may be hesitant due to uncertainty, or feel anxious due to cumbersome operations. If these emotions are not handled in a timely manner, users may interrupt operations, abandon transactions, or even make irrational decisions, ultimately leading to business loss and a decline in user experience.

[0003] The existing financial service process lacks the ability to dynamically perceive user emotions, cannot adjust the process according to the user's real-time emotional state, and cannot meet the personalized needs of users in different situations. Therefore, an artificial intelligence-based internal process control method for financial institutions is urgently needed to solve the above problems. Summary of the invention

[0004] The present invention provides an internal process control method and system for a financial institution based on artificial intelligence, which solves the technical problems in the related technology that the internal process control method and system lack the ability to dynamically perceive user emotions, cannot adjust the process according to the user's real-time emotional state, and cannot meet the user's personalized needs in different situations.

[0005] The present invention provides an artificial intelligence-based internal process control method and system for a financial institution, comprising the following steps:

[0006] Step S101, collecting the user's behavior data and text input data when using the financial software, the behavior data includes: mouse click frequency, page dwell time, keyboard input frequency and form filling speed;

[0007] Step S102, preprocessing the behavior data and the text input data to obtain a first feature sequence and a second feature sequence respectively;

[0008] Step S103, constructing an emotion recognition model according to the first feature sequence and the second feature sequence, wherein the input of the emotion recognition model is the first feature sequence and the second feature sequence, and the output is the probability value of the user's emotional state, wherein the emotional state includes: anxiety, excitement, hesitation, confusion and satisfaction;

[0009] Step S104: generating a response measure according to the probability value of the user's emotional state output by the emotion recognition model.

[0010] Furthermore, the mouse click frequency, page dwell time, keyboard input frequency and form filling speed are represented by real number codes, wherein the mouse click frequency is quantified by "number of clicks / unit time", the keyboard input frequency is quantified by "number of input characters / unit time", and the form filling speed is quantified by "total number of filled fields / total time".

[0011] Furthermore, the step of preprocessing the behavior data and text input data includes:

[0012] Step S201, cleaning the text input data, the cleaning means include: removing irrelevant characters, correcting typos and removing spaces;

[0013] Step S202, segmenting and encoding the text input data, specifically, using a word segmenter based on the BERT model to segment the text input data into words or phrases, encoding the segmented text input data using the BERT model, and representing it through a first feature sequence;

[0014] Step S203, dividing the collected behavior data into behavior data of N time points according to a preset time interval, where N is a custom parameter;

[0015] Step S204, using the standard deviation method to remove outliers in the behavior data, specifically: calculating the mean and standard deviation of the behavior data at the nth time point, and deleting the behavior data when it is outside three standard deviations of the mean, where 1≤n≤N;

[0016] Step S205, using the z-score normalization method to normalize the behavior data to obtain a second feature sequence.

[0017] Furthermore, the emotion recognition model includes: a first hidden layer, a second hidden layer, an expansion layer, a concatenation layer and a classifier;

[0018] The first hidden layer is used to input the first feature sequence and output the first feature matrix;

[0019] The second hidden layer is used to input the second feature sequence and output the second feature vector;

[0020] The expansion layer is used to expand the first feature matrix into a first feature vector, and use the principal component analysis method to perform dimensionality reduction processing on the first feature vector and the second feature vector to obtain a third feature vector and a fourth feature vector;

[0021] The concatenation layer is used to concatenate the third eigenvector and the fourth eigenvector to obtain a concatenated vector;

[0022] The concatenated vector is input into the classifier, and the classification space of the classifier represents the probability value of the user's emotional state.

[0023] Furthermore, the calculation formula of the first hidden layer is:

[0024]

[0025] Where Attention represents the first feature matrix output by the first hidden layer, Query, Key and Value represent the first intermediate feature, the second intermediate feature and the third intermediate feature of the first hidden layer respectively, T represents the transposition operation, and softmax represents the softmax activation function.

[0026] Furthermore, the second hidden layer is constructed based on a gated recurrent unit.

[0027] Furthermore, the step of performing dimensionality reduction processing on the first eigenvector and the second eigenvector using principal component analysis method includes:

[0028] Step S301, normalizing the first eigenvector and the second eigenvector;

[0029] Step S302, calculating the covariance matrix according to step S301;

[0030] Step S303, performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors;

[0031] Step S304, sorting by the size of the eigenvalues, and selecting the eigenvectors corresponding to the first k eigenvalues, where k is a custom parameter;

[0032] Step S305 , projecting the first eigenvector and the second eigenvector to a low-dimensional space through the eigenvector selected in step S304 to obtain a vector after dimensionality reduction.

[0033] Furthermore, manual annotation is used as sample labels for training samples of the emotion recognition model.

[0034] Furthermore, according to the probability value of the emotional state output by the emotion recognition model, different initial thresholds are set to trigger corresponding response measures, and the thresholds are dynamically adjusted according to the user's behavior and feedback.

[0035] The present invention provides an artificial intelligence-based internal process control system for a financial institution, comprising the following modules:

[0036] The data collection module is used to collect the user's behavior data and text input data when using the financial software;

[0037] A preprocessing module, which is used to preprocess the behavior data and the text input data to obtain a first feature sequence and a second feature sequence respectively;

[0038] An emotion recognition module, which is used to construct an emotion recognition model according to the first feature sequence and the second feature sequence, and output a probability value of the user's emotional state;

[0039] The response measure generation module is used to generate a response measure according to the user's emotional state probability value output by the emotion recognition model.

[0040] The beneficial effects of the present invention are as follows: by collecting the user's behavioral data and text input data, the present invention constructs an emotion recognition model, which can dynamically perceive the user's emotional state and analyze it in real time, thereby adaptively adjusting the operating process according to the user's emotional changes and improving the user's interactive experience. Compared with the traditional fixed process design, the system of the present invention has higher flexibility and can be dynamically adjusted according to the user's real-time situation, thereby improving the intelligence and efficiency of the process. Through this timely intervention and optimization, the system effectively reduces the risk of users interrupting operations or making inappropriate decisions, and significantly improves the success rate of financial services and business conversion rates. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flow chart of an internal process control method of a financial institution based on artificial intelligence of the present invention;

[0042] Figure 2 It is a flow chart of the present invention for preprocessing behavior data and text input data. DETAILED DESCRIPTION

[0043] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0044] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in one or more embodiments of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0045] like Figure 1-Figure 2 As shown, a method and system for controlling internal processes of a financial institution based on artificial intelligence comprises the following steps:

[0046] Step S101, collecting the user's behavior data and text input data when using the financial software, the behavior data includes: mouse click frequency, page dwell time, keyboard input frequency and form filling speed;

[0047] Step S102, preprocessing the behavior data and the text input data to obtain a first feature sequence and a second feature sequence respectively;

[0048] Step S103, constructing an emotion recognition model according to the first feature sequence and the second feature sequence, wherein the input of the emotion recognition model is the first feature sequence and the second feature sequence, and the output is the probability value of the user's emotional state, wherein the emotional state includes: anxiety, excitement, hesitation, confusion and satisfaction;

[0049] Step S104: generating a response measure according to the probability value of the user's emotional state output by the emotion recognition model.

[0050] In one embodiment of the present invention, the user's behavior data and the text input data of the user's conversation with the intelligent customer service built into the financial software are collected through a browser-based JavaScript library, and the behavior data and text input data are transmitted to the background.

[0051] In one embodiment of the present invention, the mouse click frequency, page dwell time, keyboard input frequency and form filling speed are represented by real number coding, wherein the mouse click frequency is quantified by "number of clicks / unit time (such as seconds)", the keyboard input frequency is quantified by "number of input characters / unit time (such as seconds)", and the form filling speed is quantified by "total number of filled fields / total time".

[0052] In one embodiment of the present invention, the step of preprocessing the behavior data and text input data includes:

[0053] Step S201, cleaning the text input data, the cleaning means include: removing irrelevant characters, correcting typos and removing spaces;

[0054] Step S202, segmenting and encoding the text input data, specifically, using a word segmenter based on the BERT model to segment the text input data into words or phrases, encoding the segmented text input data using the BERT model, and representing it through a first feature sequence;

[0055] Step S203, dividing the collected behavior data into behavior data of N time points according to a preset time interval, where N is a custom parameter;

[0056] Step S204, using the standard deviation method to remove outliers in the behavior data, specifically: calculating the mean and standard deviation of the behavior data at the nth time point, and deleting the behavior data when it is outside three standard deviations of the mean, where 1≤n≤N;

[0057] Step S205, using the z-score normalization method to normalize the behavior data to obtain a second feature sequence.

[0058] In one embodiment of the present invention, the emotion recognition model comprises: a first hidden layer, a second hidden layer, an expansion layer, a concatenation layer and a classifier;

[0059] The first hidden layer is used to input the first feature sequence and output the first feature matrix;

[0060] The second hidden layer is used to input the second feature sequence and output the second feature vector;

[0061] The expansion layer is used to expand the first feature matrix into a first feature vector, and use the principal component analysis method to perform dimensionality reduction processing on the first feature vector and the second feature vector to obtain a third feature vector and a fourth feature vector;

[0062] The concatenation layer is used to concatenate the third eigenvector and the fourth eigenvector to obtain a concatenated vector;

[0063] The concatenated vector is input into the classifier, and the classification space of the classifier represents the probability value of the user's emotional state.

[0064] In one embodiment of the present invention, the calculation formula of the first hidden layer is:

[0065]

[0066] Where Attention represents the first feature matrix output by the first hidden layer, Query, Key and Value represent the first intermediate feature, the second intermediate feature and the third intermediate feature of the first hidden layer respectively, T represents the transposition operation, and softmax represents the softmax activation function.

[0067] In one embodiment of the present invention, the second hidden layer is constructed based on a gated recurrent unit.

[0068] In one embodiment of the present invention, the calculation formula of the second hidden layer includes:

[0069] The calculation formula of the reset gate of the second hidden layer is:

[0070]

[0071] Among them, x n represents the nth sequence unit of the second characteristic sequence, h n-1 represents the n-1th second eigenvector, and Respectively represent the first weight parameter, the second weight parameter and the bias parameter corresponding to the reset gate of the nth sequence unit of the second feature sequence;

[0072] The calculation formula of the update gate of the third hidden layer is:

[0073]

[0074] in, and Respectively represent the first weight parameter, the second weight parameter and the bias parameter corresponding to the update gate of the nth sequence unit of the second feature sequence;

[0075] The calculation formula for the candidate hidden features of the third hidden layer is:

[0076]

[0077] in, r n and Respectively represent the first weight parameter, the second weight parameter and the bias parameter corresponding to the candidate hidden feature of the nth sequence unit of the second feature sequence;

[0078] The calculation formula for the hidden state update of the third hidden layer is:

[0079]

[0080] Among them, z n and represents the update gate and candidate hidden features of the nth sequence unit of the second feature sequence, and ⊙ represents point-by-point multiplication.

[0081] In one embodiment of the present invention, the step of performing dimensionality reduction processing on the first eigenvector and the second eigenvector using principal component analysis includes:

[0082] Step S301, normalize the first eigenvector and the second eigenvector, and the normalization calculation formula is: Where X represents the value of each feature, μ represents the mean of each feature, and σ represents the standard deviation of each vector;

[0083] Step S302, calculating the covariance matrix according to step S301, the calculation formula of the covariance matrix is: C = (X norm -μ) T (X norm -μ), where T represents the transpose operation;

[0084] Step S303, performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors, where the eigenvectors represent the principal component directions of the data, and the eigenvalues ​​represent the variance in that direction;

[0085] Step S304, sorting by the size of the eigenvalues, selecting the eigenvectors corresponding to the first k eigenvalues, where k is a custom parameter, and preferably, the first k eigenvectors selected retain 95% of the information of the first eigenvector and the second eigenvector;

[0086] Step S305, projecting the first eigenvector and the second eigenvector to a low-dimensional space through the eigenvector selected in step S304 to obtain a vector after dimensionality reduction, and the projection formula is: r =v c *W k , where v c represents the first and second eigenvectors, W k represents the feature matrix consisting of the first k eigenvectors, v r Represents the first eigenvector and the second eigenvector after dimensionality reduction.

[0087] In one embodiment of the present invention, manual annotation is used as sample labels of training samples of the emotion recognition model.

[0088] In one embodiment of the present invention, a cross entropy loss function is used as the loss function of the emotion recognition model.

[0089] In one embodiment of the present invention, different initial thresholds are set to trigger corresponding response measures according to the probability value of the emotional state output by the emotion recognition model, and the threshold is dynamically adjusted according to the user's behavior and feedback. For example, the initial threshold of anxiety is set to 0.6. When the probability value of anxiety output by the emotion recognition model exceeds 0.6, the user is considered to be in an anxious state. The response measures taken at this time are: simplifying the operation process, providing guidance information, and enabling chatbots to provide instant help; if the anxiety response is frequently triggered, the threshold can be adjusted from 0.6 to 0.7 to reduce unnecessary intervention.

[0090] In one embodiment of the present invention, there is also provided an artificial intelligence-based internal process control system for a financial institution, comprising the following modules:

[0091] The data collection module is used to collect the user's behavior data and text input data when using the financial software;

[0092] A preprocessing module, which is used to preprocess the behavior data and the text input data to obtain a first feature sequence and a second feature sequence respectively;

[0093] An emotion recognition module, which is used to construct an emotion recognition model according to the first feature sequence and the second feature sequence, and output a probability value of the user's emotional state;

[0094] The response measure generation module is used to generate a response measure according to the user's emotional state probability value output by the emotion recognition model.

[0095] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. A method for controlling internal processes of a financial institution based on artificial intelligence, characterized in that: The following steps are involved: Step S101, collecting the user's behavior data and text input data when using the financial software, the behavior data includes: mouse click frequency, page dwell time, keyboard input frequency and form filling speed; Step S102, preprocessing the behavior data and the text input data to obtain a first feature sequence and a second feature sequence respectively; Step S103, constructing an emotion recognition model according to the first feature sequence and the second feature sequence, wherein the input of the emotion recognition model is the first feature sequence and the second feature sequence, and the output is the probability value of the user's emotional state, wherein the emotional state includes: anxiety, excitement, hesitation, confusion and satisfaction; Step S104: generating a response measure according to the probability value of the user's emotional state output by the emotion recognition model.

2. According to claim 1, a method for controlling internal processes of a financial institution based on artificial intelligence is characterized in that: The mouse click frequency, page dwell time, keyboard input frequency and form filling speed are represented by real number codes, wherein the mouse click frequency is quantified by "number of clicks / unit time", the keyboard input frequency is quantified by "number of input characters / unit time", and the form filling speed is quantified by "total number of filled fields / total time".

3. According to claim 1, a method for controlling internal processes of a financial institution based on artificial intelligence is characterized in that: The step of preprocessing the behavior data and text input data comprises: Step S201, cleaning the text input data, the cleaning means include: removing irrelevant characters, correcting typos and removing spaces; Step S202, segmenting and encoding the text input data, specifically, using a word segmenter based on the BERT model to segment the text input data into words or phrases, encoding the segmented text input data using the BERT model, and representing it through a first feature sequence; Step S203, dividing the collected behavior data into behavior data of N time points according to a preset time interval, where N is a custom parameter; Step S204, using the standard deviation method to remove outliers in the behavior data, specifically: calculating the mean and standard deviation of the behavior data at the nth time point, and deleting the behavior data when it is outside three standard deviations of the mean, where 1≤n≤N; Step S205, using the z-score normalization method to normalize the behavior data to obtain a second feature sequence.

4. The method for controlling the internal process of a financial institution based on artificial intelligence according to claim 1, characterized in that: The emotion recognition model includes: a first hidden layer, a second hidden layer, an expansion layer, a splicing layer and a classifier; The first hidden layer is used to input the first feature sequence and output the first feature matrix; The second hidden layer is used to input the second feature sequence and output the second feature vector; The expansion layer is used to expand the first feature matrix into a first feature vector, and use the principal component analysis method to perform dimensionality reduction processing on the first feature vector and the second feature vector to obtain a third feature vector and a fourth feature vector; The concatenation layer is used to concatenate the third eigenvector and the fourth eigenvector to obtain a concatenated vector; The concatenated vector is input into the classifier, and the classification space of the classifier represents the probability value of the user's emotional state.

5. According to claim 4, a method for controlling internal processes of a financial institution based on artificial intelligence is characterized in that: The calculation formula of the first hidden layer is: Where Attention represents the first feature matrix output by the first hidden layer, Query, Key and Value represent the first intermediate feature, the second intermediate feature and the third intermediate feature of the first hidden layer respectively, T represents the transposition operation, and softmax represents the softmax activation function.

6. According to claim 4, a method for controlling internal processes of a financial institution based on artificial intelligence is characterized in that: The second hidden layer is constructed based on a gated recurrent unit.

7. The method for controlling the internal process of a financial institution based on artificial intelligence according to claim 4, characterized in that: The step of performing dimensionality reduction processing on the first eigenvector and the second eigenvector using the principal component analysis method comprises: Step S301, normalizing the first eigenvector and the second eigenvector; Step S302, calculating the covariance matrix according to step S301; Step S303, performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors; Step S304, sorting by the size of the eigenvalues, and selecting the eigenvectors corresponding to the first k eigenvalues, where k is a custom parameter; Step S305 , projecting the first eigenvector and the second eigenvector to a low-dimensional space through the eigenvector selected in step S304 to obtain a vector after dimensionality reduction.

8. The method for controlling the internal process of a financial institution based on artificial intelligence according to claim 1, characterized in that: Manual annotation is used as sample labels for training samples of emotion recognition models.

9. The method for controlling the internal process of a financial institution based on artificial intelligence according to claim 1, characterized in that: According to the probability value of the emotional state output by the emotion recognition model, different initial thresholds are set to trigger corresponding response measures, and the thresholds are dynamically adjusted according to the user's behavior and feedback.

10. An artificial intelligence-based internal process control system for a financial institution, characterized in that: An artificial intelligence-based internal process control method for a financial institution as described in any one of claims 1 to 9 is adopted, comprising the following modules: The data collection module is used to collect the user's behavior data and text input data when using the financial software; A preprocessing module, which is used to preprocess the behavior data and the text input data to obtain a first feature sequence and a second feature sequence respectively; An emotion recognition module, which is used to construct an emotion recognition model according to the first feature sequence and the second feature sequence, and output a probability value of the user's emotional state; The response measure generation module is used to generate a response measure according to the user's emotional state probability value output by the emotion recognition model.

Images