Operation process analysis optimization method and device, equipment and medium

Abstract

The invention relates to the technical field of data analysis, can be applied to business system platforms of financial science and technology, medical treatment and health and the like, and discloses an operation process analysis optimization method, device, equipment and medium. Acquiring text features and operation information features of each link in a predefined business process according to the multi-dimensional source data, fusing the text features and the operation information features to obtain fused features, and identifying abnormal information and start-end time of each link according to the fused features by using a pre-trained process analysis large model to obtain abnormal information and start-end time of each link; and generating an optimization suggestion according to the starting and ending time, and feeding back the optimization suggestion and the abnormal information to the seat business personnel in real time. And the work flow analysis and optimization efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of data analysis technology, and in particular to a method, device, equipment and medium for analyzing and optimizing an operation process. Background Art

[0002] In the field of financial technology, when agents handle user inquiries, financial product recommendations, loan application assistance, and other tasks, they generate a large amount of interactive data, including voice recordings of telephone conversations, system operation logs, and video recordings in remote service scenarios. However, existing technologies for processing this data have obvious limitations: voice recordings are often difficult to extract information due to environmental noise and equipment interference, and are often stored separately from operation logs and video records, making it difficult to perform correlation analysis in the time dimension; operation log formats are often inconsistent due to differences in business systems, increasing the difficulty of cross-process data integration; video records are often stored in a rough manner due to the large amount of data, making it difficult to quickly locate agent status information at key time points. This data processing method makes it difficult to accurately identify efficiency bottlenecks and irregular operations in various links of the business process. For example, it is impossible to determine whether the agent has accurately executed the risk warning process based on voice content and synchronized operation records, and it is difficult to analyze the root causes of excessive business time through multi-source data linkage, which restricts the refinement of financial service quality.

[0003] In the healthcare sector, when agents provide users with services such as outpatient registration guidance, examination report interpretation, and medical process consultation, they also generate data from voice interactions, system operations, and video communications. However, current data processing models have many shortcomings: background noise mixed in with voice data can distort patient symptom descriptions and the extraction of agent guidance information; the operation logs generated by registration systems and electronic medical record systems in different medical institutions vary in format, making them difficult to standardize and integrate, hindering the analysis of cross-platform business processes; and video recordings lack targeted keyframe extraction, making it impossible to effectively reflect the professionalism of agents when answering medical questions (such as the standardization of facial expressions and movements). Furthermore, the lack of a time synchronization mechanism for multi-source data makes it difficult to spatially and temporally correlate patient consultation content, agent operation behavior, and service status. This makes it difficult to promptly identify problems such as errors in registration process guidance and omissions of key medical information. This not only affects the patient's medical experience, but also may introduce potential medical service risks due to non-standardized processes. Summary of the Invention

[0004] The present invention provides a method, device, computer equipment and medium for analyzing and optimizing an operation process, so as to solve the problem of low efficiency of the existing operation process analysis and optimization methods on the market.

[0005] In a first aspect, a method for analyzing and optimizing an operation process is provided, comprising: Acquire voice recordings, operation logs, and video records of agents during operations to obtain multi-dimensional source data; Acquire text features and operation information features of each link in a predefined business process based on the multidimensional source data; Fusing the text feature with the operation information feature to obtain a fused feature; Using the pre-trained process analysis model to identify abnormal information and start and end times of each link based on the fusion features; An optimization suggestion is generated according to the start and end time, and the optimization suggestion and the abnormal information are fed back to the agent in real time.

[0006] In a second aspect, a device for analyzing and optimizing a work process is provided, comprising: The data acquisition module is used to obtain the voice recordings, operation logs and video records of the seat business personnel during the operation process to obtain multi-dimensional source data; A feature extraction module, configured to obtain text features and operation information features of each link in a predefined business process based on the multi-dimensional source data; A feature fusion module, configured to fuse the text feature with the operation information feature to obtain a fused feature; An anomaly identification module, used to use the pre-trained process analysis model to identify the anomaly information and start and end time of each link according to the fusion features; The optimization feedback module is used to generate optimization suggestions according to the start and end times, and to provide the agent with real-time feedback on the optimization suggestions and the abnormal information.

[0007] In a third aspect, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned job flow analysis and optimization method when executing the computer program.

[0008] In a fourth aspect, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the steps of the above-mentioned job process analysis and optimization method are implemented.

[0009] The solution implemented by the aforementioned workflow analysis and optimization method, device, computer equipment, and storage medium can acquire multidimensional source data, including agent voice recordings, operation logs, and video recordings. This approach has specific applications in both fintech (e.g., telephone consultations for financial services) and healthcare (e.g., consultations for outpatient appointments). The acquisition process requires voice denoising, log format standardization, and video frame capture at fixed intervals. These three elements are then aligned and aggregated using timestamps. Next, structured basic information is extracted from the multidimensional source data. The process is divided into stages according to pre-set business processes. Text features (e.g., keywords, semantic themes) and operational information features (e.g., operation type, efficiency) are extracted for each stage. These two types of features are then standardized and dimensionally aligned, weighted and fused using an attention mechanism, and then enhanced through feature interaction to produce fused features. The fused features are then processed using a pre-trained large-scale process analysis model (trained using online financial consultation data in the financial sector and doctor-patient conversation data in the healthcare sector). The start and end times of each stage and any anomalies are determined through encoding, high-dimensional mapping, and time anchor point screening. Finally, the time loss is analyzed based on the start and end time, and the abnormal links and causes are found by comparing with the standard time. The optimization strategy library is matched and the feasibility is verified to generate optimization suggestions. Suggestions and abnormal information are fed back to the seats in real time, which improves the efficiency of operation process analysis and optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0011] Figure 1 This is a schematic diagram of an application environment of a method for analyzing and optimizing a work process according to an embodiment of the present invention; Figure 2 This is a flow chart of a method for analyzing and optimizing an operation process according to an embodiment of the present invention; Figure 3 This is a structural diagram of an operation process analysis and optimization device according to an embodiment of the present invention; Figure 4 is a structural diagram of a computer device in one embodiment of the present invention; Figure 5 FIG. 2 is another structural diagram of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0012] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0013] The operation process analysis and optimization method provided by the embodiment of the present invention can be applied in the following Figure 1 In an application environment, the client communicates with the server over the network. The server can access multi-dimensional source data, including agent voice recordings, operation logs, and video recordings, through the client. This approach has specific applications in both FinTech (e.g., telephone consultations for financial services) and healthcare (e.g., consultations for outpatient appointments). This acquisition process requires voice denoising, log format standardization, and the acquisition of video frames at fixed intervals. These three elements are then aligned and aggregated using timestamps. Next, structured basic information is extracted from the multi-dimensional source data. The process is divided into stages according to pre-defined business processes. Text features (e.g., keywords, semantic themes) and operation information features (e.g., operation type, efficiency) are extracted for each stage. These two types of features are then standardized and dimensionally aligned, weighted and fused using an attention mechanism, and then enhanced through feature interaction to generate fused features. The fused features are then processed using a pre-trained process analysis model (trained using online financial consultation data in the financial sector and doctor-patient conversation data in the healthcare sector). Through encoding, high-dimensional mapping, and screening of time anchors, the start and end times of each stage and any anomalies are determined. Finally, time loss is analyzed based on the start and end times, and abnormal links and causes are identified by comparison with standard times. The optimization strategy library is then matched, feasibility is verified, and optimization suggestions are generated. These suggestions and abnormality information are then fed back to the agent in real time, improving the efficiency of workflow analysis and optimization. The client can include, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented as a standalone server or a server cluster consisting of multiple servers. The present invention is described in detail below using specific embodiments.

[0014] See also Figure 2 As shown, Figure 2 A flowchart of a method for analyzing and optimizing a work process according to an embodiment of the present invention includes the following steps: S1. Obtain the voice recordings, operation logs, and video records of the agent during the operation to obtain multi-dimensional source data.

[0015] In the field of financial technology, the voice recordings, operation logs and video records of the agent during the acquisition process can be the voice recordings, operation logs and video records obtained by the user when consulting financial-related services by phone.

[0016] In the medical and health field, the voice recording, operation log and video record of the seat business personnel during the acquisition operation process can be the voice recording, operation log and video record obtained by the user during the consultation outpatient registration process.

[0017] In the embodiment of the present invention, the multi-dimensional source data obtained by obtaining the voice recordings, operation logs and video records of the agent during the operation includes: Performing denoising on the voice recording to obtain denoised voice data; Performing format standardization processing on the operation log to obtain a standardized log; Collecting frame images in the video record based on a preset collection interval to obtain a frame image set; Aligning the denoised speech data, the standardized log, and the frame picture set based on the timestamp to obtain aligned denoised speech data, aligned standardized log, and aligned frame picture set; The aligned denoised speech data, the aligned normalized logs, and the aligned frame image set are aggregated to obtain multi-dimensional source data.

[0018] In detail, the voice recording is denoised to obtain denoised voice data. The original voice recording can be processed by an audio denoising algorithm (such as noise reduction technology based on spectrum analysis) to filter out irrelevant noise and retain clear human voice interaction content, ultimately obtaining denoised voice data, laying the foundation for subsequent voice feature extraction and analysis.

[0019] In detail, the format standardization processing of the operation log to obtain a standardized log can be performed according to preset format specifications (such as unified field names, timestamp formats, operation type codes, etc.) to eliminate format differences and form a structured standardized log to ensure compatibility during subsequent data fusion and analysis.

[0020] Specifically, capturing frames from the video recording at a preset interval can involve setting a fixed interval (e.g., capturing one frame per second) and extracting key frames from the video recording to form a frame image set. This reduces the amount of video data stored while allowing the frames to reflect the agent's state (e.g., facial expression, operating movements, etc.) at different points in time, meeting the requirements for subsequent video analysis.

[0021] In detail, the timestamp-based alignment of the denoised voice data, the standardized log, and the frame picture set is to accurately match the three in the time dimension through a time synchronization algorithm, ensuring that the voice, operation, and video information at the same time point correspond to each other, and obtaining the aligned denoised voice data, aligned standardized log, and aligned frame picture set, providing a time consistency basis for the fusion of multi-source data.

[0022] S2. Acquire text features and operation information features of each link in a predefined business process based on the multi-dimensional source data.

[0023] In an embodiment of the present invention, the step of obtaining text features and operation information features of each link in a predefined business process based on the multidimensional source data includes: Extracting basic information from the multidimensional source data to obtain structured basic information; Dividing the structured basic information into sections based on a preset business process to obtain a section division result; The text features and operation information of each link are extracted according to the link division result, and the operation information features are generated according to the operation information.

[0024] Specifically, extracting basic information from the multidimensional source data to obtain structured basic information involves converting denoised speech data into text using speech-to-text technology, extracting key information such as operation type, operation object, and operation time from standardized logs, and extracting visual features related to business processes (such as keyframe descriptions of agent operation actions) from frame image collections. This information is then organized into a unified data format (e.g., a structured table) to form structured basic information.

[0025] Specifically, segmenting the structured basic information based on the pre-set business process involves matching the text content, operation records, and timestamps within the structured basic information with the characteristics of the pre-set process segments. For example, by identifying the voice text "Hello, how can I help you?" as the start of the "Customer Inquiry Access" segment, and by identifying the action "Submit Order" in the operation log as the end of the "Order Processing" segment, the start and end times of each segment can be determined in combination with the timestamps, ultimately resulting in a segmentation result that clearly distinguishes each segment.

[0026] Specifically, the text features for each stage, extracted based on the stage segmentation results, are based on natural language processing. These features include core keywords (e.g., "refund," "logistics inquiry"), semantic topics (e.g., "after-sales problem handling"), and conversational intent (e.g., customer needs and agent response objectives). For example, in the "needs understanding" stage, specific customer-specific demand keywords are extracted from the text to form the text features for that stage.

[0027] Specifically, the process of extracting operational information from each stage based on the stage classification results and generating operational information features based on this operational information involves extracting specific operational information from the standardized log corresponding to each stage, including the operation type (e.g., query, entry, modification), operation object (e.g., customer information table, order system), number of operations, operation duration, and operation result (e.g., success, failure). Based on this operational information, a large-scale model is then used to analyze and generate operational information features, such as operational efficiency (average single-step operation duration, total operation duration for each stage), operational standardization (conformance to pre-set operational procedures, presence of redundant operations), and operational relevance (sequential order and logical relationships between different operations). These features objectively reflect the agent's operational behavior in that stage, providing a basis for subsequent process analysis and optimization.

[0028] S3. Fusing the text feature with the operation information feature to obtain a fused feature.

[0029] In the embodiment of the present invention, fusing the text feature with the operation information feature to obtain a fused feature includes: Performing feature standardization processing on the text features and the operation information features to obtain standardized text features and standardized operation features; Performing feature dimension alignment on the standardized text features and the standardized operation features to obtain aligned text features and aligned operation features; Calculate the weights of the text features and the operation information features in the business process based on the attention mechanism to obtain feature weights; Performing weighted fusion on the aligned text features and the aligned operation features according to the feature weights to obtain preliminary fusion features; Perform feature interaction enhancement on the preliminary fusion features to obtain fusion features.

[0030] In detail, the feature standardization processing of the text features and the operation information features is to eliminate the dimensional and format differences of the two types of features through standardization processing (such as normalization to map numerical values ​​to a unified interval, and one-hot encoding to process categorical features), so that the standardized text features and operation features have the basic conditions for fusion, ensuring that the influence of each type of features in the subsequent fusion process is in a fair and comparable dimension.

[0031] In detail, the feature dimension alignment of the standardized text features and the standardized operation features is to adjust the dimensions of the two types of features through dimensional mapping technology (such as dimensionality increase by embedding layer, dimensionality reduction by principal component analysis, etc.), so that the dimensions of the standardized text features and the standardized operation features remain consistent, and the aligned text features and operation features are obtained, providing a unified dimensional basis for subsequent weighted calculations.

[0032] In detail, the attention mechanism-based calculation of the weights of the text features and the operation information features in the business process can be to learn the correlation strength between the two types of features and business goals in different links through large model analysis technology, calculate their respective weight values ​​(such as text feature weight 0.7, operation feature weight 0.3), and form feature weights for quantifying the influence of the two types of features in the current link.

[0033] In detail, the feature interaction enhancement of the preliminary fusion features is to mine the deep correlation between text features and operation information features through feature interaction technology (such as element-level multiplication, splicing followed by neural network mapping, etc.), strengthen the synergistic effect between features, so that the final fusion features not only contain independent information of the two types of features, but also reflect the interactive relationship between the two in the business process, providing more comprehensive and accurate feature support for subsequent process analysis and anomaly identification.

[0034] S4. Using the pre-trained process analysis model, identify the abnormal information and start and end time of each link according to the fusion features.

[0035] In the fintech sector, large process analysis models can be built by collecting massive amounts of real-world conversation records from online financial consulting platforms, such as questions from clients about the returns and risks of financial products, and the consultants' responses. This data reflects the realities of consulting services and can help the model learn how to understand client questions and provide appropriate responses. It can also be used to train the model's ability to handle complex questions, multi-round conversations, and ambiguous or incomplete questions, allowing the model to better adapt to the financial context.

[0036] In the healthcare sector, the large process analysis model can be trained by collecting a large amount of doctor-patient conversation data from online medical consultation platforms. These conversations reflect real-world scenarios, such as patient descriptions of symptoms, doctor inquiries, diagnostic processes, and advice given. This helps the model learn doctor-patient communication patterns, problem diagnosis ideas, and response strategies. For example, approximately 100,000 interactions collected from HealthCareMagic and approximately 10,000 conversations from iCliniq were used to train the ChatDoctor model.

[0037] In the embodiment of the present invention, the use of the pre-trained process analysis model to identify abnormal information and start and end times of each link based on the fusion features includes: Using the process analysis model to perform deep encoding on the fusion features to obtain encoding features; Performing high-dimensional feature mapping on the coding feature to obtain a high-dimensional feature vector; Filter candidate time anchor points in the high-dimensional feature vector that may correspond to the start and end of a link based on the timestamp association information implicit in the fusion feature; Based on the standard time range of each link included in the preset business process, the candidate time anchor points are eliminated as abnormal anchor points to obtain screening anchor points; Confirm the start and end time of each link based on the screening anchor point; The abnormal links in the high-dimensional feature vector are detected according to the above-mentioned start and end times and the preset link abnormality judgment rules to obtain abnormal information.

[0038] Specifically, deep encoding of the fused features using the process analysis model involves deep encoding the fused features using a multi-layer neural network (such as a Transformer architecture). This process abstracts the semantic information, operational logic, and temporal relationships in the original fused features into structured encoded features. This process is equivalent to "information compression and refinement" of the fused features, preserving core features while reducing data redundancy, providing more accurate input for subsequent high-dimensional feature mapping and analysis.

[0039] Specifically, high-dimensional feature mapping of the encoded features involves mapping them to a higher-dimensional space through the feature mapping layer of the large model (such as a fully connected layer or embedding layer), generating a high-dimensional feature vector. High-dimensional feature vectors can more carefully capture subtle differences implicit in the fused features (such as subtle differences in operating habits at different stages, emotional changes during voice interaction, etc.), thereby improving the accuracy of subsequent recognition of process details and providing richer feature support for time anchor point screening and anomaly detection.

[0040] Specifically, the method of screening candidate time anchor points in the high-dimensional feature vector that may correspond to the start and end of a link based on the timestamp association information implicit in the fused features utilizes this timestamp association information, combined with the temporal variation patterns of the high-dimensional feature vector, to screen out time points where the feature distribution in the high-dimensional feature vector undergoes a significant abrupt change. These are then used as candidate time anchor points that may correspond to the start or end of a link. These candidate anchor points initially mark possible boundaries between process links, laying the foundation for the subsequent precise determination of link start and end points.

[0041] In detail, the abnormal anchor point elimination of the candidate time anchor points based on the standard time range of each link included in the preset business process is to compare the time interval corresponding to the candidate time anchor point with the standard time range of each link. If the time interval corresponding to a candidate anchor point is far higher or lower than the standard range (for example, the interval between the candidate anchor points of a link is only 1 minute, which is far lower than the standard range of 3 minutes), it is determined to be an abnormal anchor point and is eliminated.

[0042] Specifically, the start and end times of each stage, determined based on the screening anchor points, are determined by combining the distribution differences of high-dimensional feature vectors (e.g., significant semantic and operational logic differences between high-dimensional feature vectors corresponding to adjacent screening anchor points) to further verify the accuracy of the screening anchor points. For example, if the high-dimensional feature vectors before and after a screening anchor point correspond to the characteristic patterns of "customer consultation" and "solution recommendation," respectively, then that anchor point can be confirmed as the dividing point between the two stages. This in turn clarifies the specific start and end times of each stage, establishing clear time boundaries for each stage.

[0043] Specifically, detecting abnormal links in the high-dimensional feature vector based on the aforementioned start and end times and pre-set link anomaly determination rules involves extracting the corresponding high-dimensional feature vector for each link and performing a comparative analysis based on pre-set link anomaly determination rules (e.g., operation duration exceeding the standard range by 20%, missing key operation steps, and the presence of illegal language in voice interaction). For example, if the high-dimensional feature vector for the "Risk Assessment" link indicates a missing "Customer Risk Level Confirmation" record in the operation log and meets the "missing key operation" condition in the anomaly determination rules, the link is determined to be abnormal, and information such as the anomaly type and occurrence time is recorded. Ultimately, a list of abnormal information is generated, providing a concrete basis for the subsequent generation of optimization recommendations.

[0044] S5. Generate optimization suggestions based on the start and end times, and provide the agent with real-time feedback on the optimization suggestions and the abnormal information.

[0045] In the field of financial technology, the abnormal information may refer to the agent's failure to provide key information such as risk level and rate of return calculation method as required when recommending financial products; the omission of necessary operational steps such as customer credit inquiry and income proof verification in the loan application review process; and the incorrect or missing filling of core data such as ID number and bank card number when entering customer information.

[0046] In the medical and health field, the abnormal information may refer to the agent guiding the customer to the wrong department during the outpatient registration process (such as directing a cardiology disease to the gastroenterology department); the registration time and treatment process informed are inconsistent with the actual regulations of the hospital; the precautions for the examination items (such as fasting requirements, contraindications to drugs) are explained incorrectly or omitted, etc.

[0047] In the embodiment of the present invention, generating the optimization suggestion according to the start and end times includes: Extract the time loss of each link according to the start and end time to obtain a time loss table; Comparing the time loss table with a preset standard time range table to obtain a comparison result; Identify abnormal links according to the comparison results, analyze abnormal causes of the abnormal links according to the abnormal links, and obtain abnormal cause analysis results; Matching the abnormality cause analysis results with a preset optimization strategy library to obtain preliminary optimization suggestions; The feasibility of the preliminary optimization suggestion is verified, and the preliminary optimization suggestion is confirmed according to the verification result to obtain the final optimization suggestion.

[0048] In detail, the time loss of each link is extracted based on the start and end time by calculating the actual time consumed by each link (end time minus start time) and combining it with the time efficiency benchmark implicit in the business process (such as the theoretical time consumed when there are no redundant operations) to extract the time loss caused by poor process, redundant operations, etc. in each link.

[0049] In detail, the comparison of the time loss table with the preset standard time range table is to compare the actual time consumption and time loss value of each link in the time loss table with the standard time range table item by item, to clarify which links have actual time consumption exceeding the upper limit of the standard, time loss value too large, or actual time consumption lower than the lower limit of the standard (there may be operational omissions), and finally form a comparison result including link identification, standard time range, actual data, and difference value, which intuitively presents the time efficiency deviation of each link.

[0050] Specifically, the process of identifying abnormal links based on the comparison results and analyzing the causes of these abnormal links involves identifying, based on the comparison results, links with actual processing time exceeding the standard range and exhibiting abnormal time loss as abnormal links. For these abnormal links, the specific causes of time loss are analyzed by combining information such as system-recorded operation logs (e.g., repeated operations, omitted steps) and voice interaction content (e.g., repeated explanations due to miscommunication). For example, if the "order submission" step takes an excessively long time, and operation logs reveal multiple submission failures, the cause can be determined to be "system response delays leading to repeated submissions." If the "requirement confirmation" step consumes significant time, and voice recordings reveal vague customer requirements, this can be attributed to "inadequate prior inquiry leading to supplemental information collection." Ultimately, an analysis of the abnormalities is generated, encompassing the abnormal link, the loss manifestations, and the specific causes.

[0051] Specifically, the abnormal cause analysis result is matched with a preset optimization strategy library. In the embodiment of the present invention, the real-time feedback of the optimization suggestions and the abnormal information to the seat business personnel is to accurately match the specific causes in the abnormal cause analysis results with the entries in the optimization strategy library, match corresponding solutions for each abnormal link, and form preliminary optimization suggestions.

[0052] Specifically, the feasibility of the preliminary optimization suggestions is verified, and the preliminary optimization suggestions are confirmed based on the verification results to obtain the final optimization suggestions. The feasibility of the preliminary optimization suggestions must be verified in conjunction with actual business scenarios. For example, the feasibility of the suggestions must be determined to determine whether they conform to the existing system architecture (e.g., whether "optimizing the system interface" is technically feasible), whether they will increase the burden on other links (e.g., whether "adding information verification steps" will make the overall process lengthy), and whether they are consistent with personnel operating habits (e.g., whether "changing the operation path" will require extensive training). Verification is conducted by referencing the results of historical optimization cases and soliciting feedback from business personnel. Infeasible or overly risky suggestions are eliminated, and the retained suggestions are further refined with regard to the operational steps and expected results, ultimately forming final, implementable optimization suggestions.

[0053] In the embodiment of the present invention, the providing of real-time feedback of the optimization suggestion and the abnormal information to the agent includes: Display the abnormal information to the agent based on a system pop-up window; Converting the optimization suggestion into audio data to obtain optimization suggestion audio; Play the optimization suggestion audio to the agent in real time.

[0054] As can be seen, in the above solution, multi-dimensional source data, including agent voice recordings, operation logs, and video recordings, is acquired. This data has specific applications in both FinTech (e.g., telephone consultations for financial services) and healthcare (e.g., consultations for outpatient appointments). The acquisition process requires voice denoising, log format standardization, and video frame capture at regular intervals. These three elements are then timestamped and aggregated. Next, structured basic information is extracted from the multi-dimensional source data. The business process is divided into stages according to pre-defined processes. Text features (e.g., keywords, semantic themes) and operation information features (e.g., operation type, efficiency) are extracted for each stage. These two types of features are then normalized and dimensionally aligned. Weights are calculated using an attention mechanism, followed by weighted fusion. Fusion features are then generated through feature interaction enhancement. The fused features are then processed using a pre-trained process analysis model (trained using online financial consultation data in the financial sector and doctor-patient conversation data in the healthcare sector). Through encoding, high-dimensional mapping, and screening of time anchors, the start and end times of each stage and any anomalies are determined. Finally, the time loss is analyzed based on the start and end time, and the abnormal links and causes are found by comparing with the standard time. The optimization strategy library is matched and the feasibility is verified to generate optimization suggestions. Suggestions and abnormal information are fed back to the seats in real time, which improves the efficiency of operation process analysis and optimization.

[0055] It should be understood that the order of execution of the steps in the above embodiments does not necessarily mean the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0056] In one embodiment, a work process analysis and optimization device is provided, which corresponds one-to-one to the work process analysis and optimization method in the above embodiment. Figure 3 As shown, the operation process analysis and optimization device includes a data acquisition module 101, a feature extraction module 102, a feature fusion module 103, an anomaly identification module 104, and an optimization feedback module 105. The detailed description of each functional module is as follows: The data acquisition module 101 is used to acquire the voice recordings, operation logs and video records of the seat business personnel during the operation process to obtain multi-dimensional source data; A feature extraction module 102 is configured to obtain text features and operation information features of each link in a predefined business process based on the multi-dimensional source data; A feature fusion module 103 is configured to fuse the text feature with the operation information feature to obtain a fused feature; Anomaly identification module 104, used to use the pre-trained process analysis model to identify the anomaly information and start and end time of each link according to the fusion features; The optimization feedback module 105 is configured to generate optimization suggestions based on the start and end times, and provide the optimization suggestions and the abnormal information to the agent in real time.

[0057] In one embodiment, the data acquisition module 101, when performing the acquisition operation to obtain multi-dimensional source data from the voice recordings, operation logs, and video records of the agent, is specifically configured to: Performing denoising on the voice recording to obtain denoised voice data; Performing format standardization processing on the operation log to obtain a standardized log; Collecting frame images in the video record based on a preset collection interval to obtain a frame image set; Aligning the denoised speech data, the standardized log, and the frame picture set based on the timestamp to obtain aligned denoised speech data, aligned standardized log, and aligned frame picture set; The aligned denoised speech data, the aligned normalized logs, and the aligned frame image set are aggregated to obtain multi-dimensional source data.

[0058] In one embodiment, the feature extraction module 102, when performing the step of acquiring text features and operation information features of each link in the predefined business process based on the multi-dimensional source data, is specifically configured to: Extracting basic information from the multidimensional source data to obtain structured basic information; Dividing the structured basic information into sections based on a preset business process to obtain a section division result; The text features and operation information of each link are extracted according to the link division result, and the operation information features are generated according to the operation information.

[0059] In one embodiment, the feature fusion module 103, when performing the fusion of the text feature and the operation information feature to obtain the fused feature, is specifically configured to: Performing feature standardization processing on the text features and the operation information features to obtain standardized text features and standardized operation features; Performing feature dimension alignment on the standardized text features and the standardized operation features to obtain aligned text features and aligned operation features; Calculate the weights of the text features and the operation information features in the business process based on the attention mechanism to obtain feature weights; Performing weighted fusion on the aligned text features and the aligned operation features according to the feature weights to obtain preliminary fusion features; Perform feature interaction enhancement on the preliminary fusion features to obtain fusion features.

[0060] In one embodiment, the anomaly identification module 104, when executing the method of using the pre-trained process analysis model to identify the anomaly information and start and end time of each link according to the fusion features, is specifically configured to: Using the process analysis model to perform deep encoding on the fusion features to obtain encoding features; Performing high-dimensional feature mapping on the coding feature to obtain a high-dimensional feature vector; Filter candidate time anchor points in the high-dimensional feature vector that may correspond to the start and end of a link based on the timestamp association information implicit in the fusion feature; Based on the standard time range of each link included in the preset business process, the candidate time anchor points are eliminated as abnormal anchor points to obtain screening anchor points; Confirm the start and end time of each link based on the screening anchor point; The abnormal links in the high-dimensional feature vector are detected according to the above-mentioned start and end times and the preset link abnormality judgment rules to obtain abnormal information.

[0061] In one embodiment, the optimization feedback module 105, when executing the step of generating the optimization suggestion according to the start and end times, is specifically configured to: Extract the time loss of each link according to the start and end time to obtain a time loss table; Comparing the time loss table with a preset standard time range table to obtain a comparison result; Identify abnormal links according to the comparison results, analyze abnormal causes of the abnormal links according to the abnormal links, and obtain abnormal cause analysis results; Matching the abnormality cause analysis results with a preset optimization strategy library to obtain preliminary optimization suggestions; The feasibility of the preliminary optimization suggestion is verified, and the preliminary optimization suggestion is confirmed according to the verification result to obtain the final optimization suggestion.

[0062] In one embodiment, the optimization feedback module 105, when performing the real-time feedback of the optimization suggestion and the abnormal information to the agent, is specifically configured to: Display the abnormal information to the agent based on a system pop-up window; Converting the optimization suggestion into audio data to obtain optimization suggestion audio; Play the optimization suggestion audio to the agent in real time.

[0063] This invention provides a workflow analysis and optimization device that acquires multidimensional source data, including agent voice recordings, operation logs, and video recordings. This device has specific applications in the fields of fintech (e.g., telephone consultations for financial services) and healthcare (e.g., consultations for outpatient appointments). The acquisition process requires voice denoising, log format standardization, and video frame capture at fixed intervals. These three elements are then timestamped and aggregated. Next, structured basic information is extracted from the multidimensional source data, which is then divided into stages according to pre-set business processes. Text features (e.g., keywords, semantic themes) and operational information features (e.g., operation type, efficiency) for each stage are extracted. These two types of features are then standardized and dimensionally aligned, weighted and fused using an attention mechanism, and then enhanced through feature interaction to produce fused features. The fused features are then processed using a pre-trained process analysis model (trained using online financial consultation data in the financial sector and doctor-patient conversation data in the healthcare sector). The start and end times of each stage and any anomalies are determined through encoding, high-dimensional mapping, and time anchor point screening. Finally, the time loss is analyzed based on the start and end time, and the abnormal links and causes are found by comparing with the standard time. The optimization strategy library is matched and the feasibility is verified to generate optimization suggestions. Suggestions and abnormal information are fed back to the seats in real time, which improves the efficiency of operation process analysis and optimization.

[0064] For the specific definition of the operation process analysis and optimization device, please refer to the definition of the operation process analysis and optimization method above, which will not be repeated here. The various modules in the above-mentioned operation process analysis and optimization device can be implemented in whole or in part by software, hardware and their combination. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0065] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 4 As shown. The computer device includes a processor, memory, network interface and database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client via a network connection. When the computer program is executed by the processor, it realizes the functions or steps on the server side of a job process analysis and optimization method.

[0066] In one embodiment, a computer device is provided. The computer device may be a client, and its internal structure diagram may be as follows: Figure 5As shown. The computer device includes a processor, memory, network interface, display screen, and input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the functions or steps on the client side of a method for analyzing and optimizing a work process.

[0067] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following steps are performed: Acquire voice recordings, operation logs, and video records of agents during operations to obtain multi-dimensional source data; Acquire text features and operation information features of each link in a predefined business process based on the multidimensional source data; Fusing the text feature with the operation information feature to obtain a fused feature; Using the pre-trained process analysis model to identify abnormal information and start and end times of each link based on the fusion features; An optimization suggestion is generated according to the start and end time, and the optimization suggestion and the abnormal information are fed back to the agent in real time.

[0068] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented: Acquire voice recordings, operation logs, and video records of agents during operations to obtain multi-dimensional source data; Acquire text features and operation information features of each link in a predefined business process based on the multidimensional source data; Fusing the text feature with the operation information feature to obtain a fused feature; Using the pre-trained process analysis model to identify abnormal information and start and end times of each link based on the fusion features; An optimization suggestion is generated according to the start and end time, and the optimization suggestion and the abnormal information are fed back to the agent in real time.

[0069] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can be found in the relevant descriptions of the server side and the client side in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.

[0070] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0071] Those skilled in the art will clearly understand that for the sake of convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0072] Finally, it should be noted that if software tools or components other than those of our company appear in the application examples, they are merely for illustration and do not represent actual use. The above-described embodiments are intended only to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above-described embodiments or replace some of the technical features therein with equivalents. Such modifications or replacements do not deviate from the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention and should be included within the scope of protection of the present invention.

Claims

1. A method for analyzing and optimizing an operation process, characterized in that: include: Acquire voice recordings, operation logs, and video records of agents during operations to obtain multi-dimensional source data; Acquire text features and operation information features of each link in a predefined business process based on the multidimensional source data; Fusing the text feature with the operation information feature to obtain a fused feature; Using the pre-trained process analysis model to identify abnormal information and start and end times of each link based on the fusion features; An optimization suggestion is generated according to the start and end time, and the optimization suggestion and the abnormal information are fed back to the agent in real time.

2. The method for analyzing and optimizing the operation process according to claim 1, wherein: The multi-dimensional source data obtained during the acquisition process includes: Performing denoising on the voice recording to obtain denoised voice data; Performing format standardization processing on the operation log to obtain a standardized log; Collecting frame images in the video record based on a preset collection interval to obtain a frame image set; Aligning the denoised speech data, the standardized log, and the frame picture set based on the timestamp to obtain aligned denoised speech data, aligned standardized log, and aligned frame picture set; The aligned denoised speech data, the aligned normalized logs, and the aligned frame image set are aggregated to obtain multi-dimensional source data.

3. The method for analyzing and optimizing the operation process according to claim 1, wherein: The step of obtaining text features and operation information features of each link in a predefined business process based on the multi-dimensional source data includes: Extracting basic information from the multidimensional source data to obtain structured basic information; Dividing the structured basic information into sections based on a preset business process to obtain a section division result; The text features and operation information of each link are extracted according to the link division result, and the operation information features are generated according to the operation information.

4. The method for analyzing and optimizing the operation process according to claim 1, wherein: The fusing of the text feature with the operation information feature to obtain a fused feature includes: Performing feature standardization processing on the text features and the operation information features to obtain standardized text features and standardized operation features; Performing feature dimension alignment on the standardized text features and the standardized operation features to obtain aligned text features and aligned operation features; Calculate the weights of the text features and the operation information features in the business process based on the attention mechanism to obtain feature weights; Performing weighted fusion on the aligned text features and the aligned operation features according to the feature weights to obtain preliminary fusion features; Perform feature interaction enhancement on the preliminary fusion features to obtain fusion features.

5. The method for analyzing and optimizing the operation process according to claim 1, wherein: The pre-trained process analysis model is used to identify abnormal information and start and end times of each link based on the fusion features, including: Using the process analysis model to perform deep encoding on the fusion features to obtain encoding features; Performing high-dimensional feature mapping on the coding feature to obtain a high-dimensional feature vector; Filter candidate time anchor points in the high-dimensional feature vector that may correspond to the start and end of a link based on the timestamp association information implicit in the fusion feature; Based on the standard time range of each link included in the preset business process, the candidate time anchor points are eliminated as abnormal anchor points to obtain screening anchor points; Confirm the start and end time of each link based on the screening anchor point; The abnormal links in the high-dimensional feature vector are detected according to the above-mentioned start and end times and the preset link abnormality judgment rules to obtain abnormal information.

6. The method for analyzing and optimizing the operation process according to claim 1, wherein: Generating the optimization suggestion according to the start and end times includes: Extract the time loss of each link according to the start and end time to obtain a time loss table; Comparing the time loss table with a preset standard time range table to obtain a comparison result; Identify abnormal links according to the comparison results, analyze abnormal causes of the abnormal links according to the abnormal links, and obtain abnormal cause analysis results; Matching the abnormality cause analysis results with a preset optimization strategy library to obtain preliminary optimization suggestions; The feasibility of the preliminary optimization suggestion is verified, and the preliminary optimization suggestion is confirmed according to the verification result to obtain the final optimization suggestion.

7. The method for analyzing and optimizing the operation process according to claim 1, wherein: The real-time feedback of the optimization suggestions and the abnormal information to the agent business personnel includes: Display the abnormal information to the agent based on a system pop-up window; Converting the optimization suggestion into audio data to obtain optimization suggestion audio; Play the optimization suggestion audio to the agent in real time.

8. A device for analyzing and optimizing an operation process, characterized in that: include: The data acquisition module is used to obtain the voice recordings, operation logs and video records of the seat business personnel during the operation process to obtain multi-dimensional source data; A feature extraction module, configured to obtain text features and operation information features of each link in a predefined business process based on the multi-dimensional source data; A feature fusion module, configured to fuse the text feature with the operation information feature to obtain a fused feature; An anomaly identification module, used to use the pre-trained process analysis model to identify the anomaly information and start and end time of each link according to the fusion features; The optimization feedback module is used to generate optimization suggestions according to the start and end times, and to provide the agent with real-time feedback on the optimization suggestions and the abnormal information.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the job flow analysis and optimization method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the job flow analysis and optimization method according to any one of claims 1 to 7 are implemented.

Images