APP software use feedback data fusion method based on hierarchical matching and interaction analysis

Abstract

The invention discloses an APP software use feedback data fusion method based on hierarchical matching and interaction analysis. The method comprises the following steps: preprocessing APP software use feedback data; obtaining user use feedback level information and user use feedback interaction information according to the preprocessed APP software use feedback data; according to the user use feedback level information, performing level matching on a user comment subject to obtain a level matching result; according to the user use feedback interaction information, performing interaction analysis on the user comment state to obtain an interaction classification result; and according to a hierarchical matching result of the user comment main body and an interactive classification result of the user comment state, constructing an APP software use feedback fusion network. According to the method, user comments and user operation use feedback data are effectively fused through hierarchical matching and interactive analysis, and more accurate user experience analysis and problem positioning can be provided for developers.

Description

Technical Field

[0001] The present invention relates to an APP software usage feedback data fusion method based on hierarchical matching and interactive analysis, and belongs to the field of APP software data analysis. Background Art

[0002] With the rapid development of the mobile application market, user feedback data for APP software has gradually become an important basis for developers to optimize their products. This paper focuses on two types of user feedback data: user comments and user operations. Among them, user comments usually come from the application market and are formatted as short texts, reflecting the user's subjective feelings about the APP software; user operation data comes from the user's operational behavior during the use of the software, recording the user's specific operational behavior in the APP, such as clicking buttons and browsing interfaces. These two types of data describe the user's usage experience from a subjective and objective perspective, respectively, and have important analytical value.

[0003] However, the current fusion analysis of user reviews and user operation data presents the following major problems: First, the hierarchy of user review subjects varies. Some reviews evaluate the entire application, while others focus on a specific function, making it difficult to directly link user review objects to corresponding operational processes. Second, the interface elements in app software have varying degrees of interactivity. Some provide active user interaction (such as buttons and text boxes), while others provide passive response functions such as loading interfaces or prompts, which may correspond to different user reviews. Therefore, how to effectively fuse user reviews and user operation data to construct a comprehensive app software usage feedback network has become an urgent problem to be solved. Summary of the Invention

[0004] The present invention provides an APP software usage feedback data fusion method based on hierarchical matching and interactive analysis, aiming to effectively fuse user comments and user operation usage feedback data through hierarchical matching and interactive analysis, thereby constructing an APP software usage feedback network and providing developers with more accurate user experience analysis and problem location.

[0005] The technical solution of the present invention is:

[0006] According to a first aspect of the present invention, a method for fusing APP software usage feedback data based on hierarchical matching and interaction analysis is provided, comprising the following steps: preprocessing the APP software usage feedback data; obtaining user usage feedback hierarchical information and user usage feedback interaction information based on the preprocessed APP software usage feedback data; performing hierarchical matching on user comment subjects based on the user usage feedback hierarchical information to obtain a hierarchical matching result; performing interaction analysis on user comment status based on the user usage feedback interaction information to obtain an interaction classification result; and constructing an APP software usage feedback fusion network based on the hierarchical matching result of the user comment subjects and the interaction classification result of the user comment status.

[0007] Furthermore, the APP software uses feedback data including user comment data and user operation data, cleans the user comment data to obtain cleaned user comment data; and filters the user operation data to obtain filtered user operation data.

[0008] Furthermore, the method of obtaining user usage feedback hierarchical information and user usage feedback interaction information based on the preprocessed APP software usage feedback data includes: extracting the user comment subject and user comment status from the cleaned user comment data in the preprocessed APP software usage feedback data based on the first language model; classifying the "APP software interface component" field in the filtered user operation data in the preprocessed APP software usage feedback data to obtain the APP software interface component classification; using a word segmentation tool to segment the "APP software interface component text" field in the filtered user operation data, and extracting the segmented verbs as APP software interface component function supplementary information; using the second language model to infer the APP software interface activities in the filtered user operation data to generate APP software interface activity function supplementary information; using the obtained user comment subject, APP software interface component function supplementary information, and APP software interface activity function supplementary information as user usage feedback hierarchical information; using the obtained user comment status and APP software interface component classification as user usage feedback interaction information.

[0009] Furthermore, based on the user feedback hierarchical information, hierarchical matching is performed on the user comment subject to obtain a hierarchical matching result, including: using an embedding model to embed the user feedback hierarchical information and generate corresponding embedding vectors: comment subject embedding vector, APP software interface component function supplementary information embedding vector, APP software interface activity function supplementary information embedding vector; based on the comment subject embedding vector and the APP software interface activity function supplementary information embedding vector, the similarity between the user comment subject and the APP software interface activity function supplementary information is calculated; based on the comment subject embedding vector and the APP software interface component function supplementary information embedding vector, the similarity between the user comment subject and the APP software interface component function supplementary information is calculated; traversing each user comment subject For the current user review subject, find the maximum similarities SimSubjectActivityMax and SimSubjecWidgetMax between it and the APP software interface activity and APP software interface component; compare SimSubjectActivityMax and SimSubjecWidgetMax: if SimSubjectActivityMax is greater than SimSubjecWidgetMax, match the current user review subject with the APP software interface activity with the greatest similarity as the hierarchical matching result of the current user review subject; otherwise, match the current user review subject with the APP software interface component with the greatest similarity as the hierarchical matching result of the current user review subject.

[0010] Furthermore, based on the user usage feedback interaction information words, the user comment status is interactively analyzed to obtain interaction classification results, including: designing user comment status keywords; using an embedding model to embed the user comment status and user comment status keywords in the user usage feedback interaction information, respectively, to generate corresponding embedding vectors: a comment status embedding vector and a comment status keyword embedding vector; based on the comment status embedding vector and the comment status keyword embedding vector, calculating the similarity between the user comment status and the user comment status keyword; traversing each user comment status, for the current user comment status, finding the user comment status keyword with the highest similarity to it, and judging the user status keyword category with the highest similarity to the current user comment status: if it belongs to the interactive category, then the interactive category matching the current user status is interactive; otherwise, if it belongs to the non-interactive category, then the interactive category matching the current user status is non-interactive.

[0011] Furthermore, the method constructs an APP software usage feedback fusion network based on the hierarchical matching results of the user review subject and the interactive classification results of the user review status, including: defining the node categories of the fusion network, the node categories include user review subject nodes, user review status nodes, APP software interface activity nodes, and APP software interface component nodes; defining the edge categories of the fusion network, the edge categories include user review subject-user review status edges, user review subject-APP software interface activity edges, user review subject-APP software interface component edges, user review status-APP software interface component edges, and APP software interface activity-APP software interface component edges; creating the user review data after cleaning. There is a user comment subject-user comment status edge; based on the filtered user operation data, all APP software interface activities-APP software interface component edges are created; the hierarchical matching results of the user comment subject are traversed, and the hierarchical matching results of the current user comment subject are judged: if the hierarchical matching result of the current user comment subject is an APP software interface activity node, then a user comment subject-APP software interface activity edge is constructed; then, based on the user comment subject-user comment status edge, the interactive classification results of the user comment status corresponding to the hierarchical matching result of the current user comment subject are traversed to determine the user comment status-APP software interface component edge; otherwise, a user comment subject-APP software interface component edge is constructed.

[0012] Furthermore, the interactive classification results of the user comment status corresponding to the hierarchical matching results of the current user comment subject are traversed to determine the user comment status-APP software interface component edge, including: based on the user comment subject-user comment status edge, the interactive classification results of the user comment status corresponding to the hierarchical matching results of the current user comment subject are traversed, and for the interactive classification results of the current user comment status, candidate user comment status-APP software interface component edges are found based on the transmission process of the user comment subject-user comment status edge, the user comment subject-APP software interface activity edge and the APP software interface activity-APP software interface component edge; if the interactive classification result of the current user comment status is an interactive category, the user comment status-APP software interface component edges of non-interactive categories are filtered out from the candidate user comment status-APP software interface component edges according to the APP software interface component classification in the user usage feedback interaction information; otherwise, if it is a non-interactive category, the user comment status-APP software interface component edges of interactive categories are filtered out from the candidate user comment status-APP software interface component edges according to the APP software interface component classification.

[0013] According to the second aspect of the present invention, a device for fusion of APP software usage feedback data based on hierarchical matching and interactive analysis is provided, comprising a module of any one of the above-mentioned methods for fusion of APP software usage feedback data based on hierarchical matching and interactive analysis.

[0014] According to a third aspect of the present invention, a processor is provided, which is used to run a program, wherein when the program is running, any one of the above-mentioned APP software usage feedback data fusion methods based on hierarchical matching and interaction analysis is executed.

[0015] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, which includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the above-mentioned APP software usage feedback data fusion methods based on hierarchical matching and interactive analysis.

[0016] The beneficial effects of the present invention are:

[0017] This method first preprocesses user comment and action data to facilitate hierarchical matching and interaction analysis. Next, based on the preprocessed app usage feedback data, it analyzes the associations between heterogeneous feedback data from two dimensions: hierarchy and interaction. Furthermore, combining the latest embedding models and similarity calculation methods, it analyzes the associations between user comment and action data from different perspectives, helping to improve the accuracy of association analysis. Finally, by defining node and edge categories in the app usage feedback network and linking corresponding nodes based on similarity and interaction categories, a comprehensive app usage feedback network is constructed, providing developers with more accurate user experience analysis and problem location. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is the overall flow chart of the present invention;

[0019] Figure 2 This is the hierarchical matching method process of the user comment subject in Step 3 of the optional embodiment of the present invention;

[0020] Figure 3 This is the interactive analysis method process of the user comment status in Step 3 of the optional embodiment of the present invention. DETAILED DESCRIPTION

[0021] To make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. It should be noted that, in the absence of conflict, the embodiments in this application and the features in the embodiments can be combined with each other in any way.

[0022] Example 1:

[0023] like Figure 1-Figure 3 As shown, according to the first aspect of an embodiment of the present invention, a method for fusing APP software usage feedback data based on hierarchical matching and interactive analysis is provided, comprising the following steps: preprocessing the APP software usage feedback data; obtaining user usage feedback hierarchical information and user usage feedback interactive information based on the preprocessed APP software usage feedback data; performing hierarchical matching on the user comment subject based on the user usage feedback hierarchical information to obtain a hierarchical matching result; performing interactive analysis on the user comment status based on the user usage feedback interactive information to obtain an interactive classification result; and constructing an APP software usage feedback fusion network based on the hierarchical matching result of the user comment subject and the interactive classification result of the user comment status.

[0024] Furthermore, the APP software uses feedback data including user comment data and user operation data, cleans the user comment data to obtain cleaned user comment data; and filters the user operation data to obtain filtered user operation data.

[0025] Furthermore, the method of obtaining user usage feedback hierarchical information and user usage feedback interaction information based on the preprocessed APP software usage feedback data includes: extracting the user comment subject and user comment status from the cleaned user comment data in the preprocessed APP software usage feedback data based on the first language model; classifying the "APP software interface component" field in the filtered user operation data in the preprocessed APP software usage feedback data to obtain the APP software interface component classification; using a word segmentation tool to segment the "APP software interface component text" field in the filtered user operation data, and extracting the segmented verbs as APP software interface component function supplementary information; using the second language model to infer the APP software interface activities in the filtered user operation data to generate APP software interface activity function supplementary information; using the obtained user comment subject, APP software interface component function supplementary information, and APP software interface activity function supplementary information as user usage feedback hierarchical information; using the obtained user comment status and APP software interface component classification as user usage feedback interaction information.

[0026] Furthermore, based on the user feedback hierarchical information, hierarchical matching is performed on the user comment subject to obtain a hierarchical matching result, including: using an embedding model to embed the user feedback hierarchical information and generate corresponding embedding vectors: comment subject embedding vector, APP software interface component function supplementary information embedding vector, APP software interface activity function supplementary information embedding vector; based on the comment subject embedding vector and the APP software interface activity function supplementary information embedding vector, the similarity between the user comment subject and the APP software interface activity function supplementary information is calculated; based on the comment subject embedding vector and the APP software interface component function supplementary information embedding vector, the similarity between the user comment subject and the APP software interface component function supplementary information is calculated; traversing each user comment subject For the current user review subject, find the maximum similarities SimSubjectActivityMax and SimSubjecWidgetMax between it and the APP software interface activity and APP software interface component; compare SimSubjectActivityMax and SimSubjecWidgetMax: if SimSubjectActivityMax is greater than SimSubjecWidgetMax, match the current user review subject with the APP software interface activity with the greatest similarity as the hierarchical matching result of the current user review subject; otherwise, match the current user review subject with the APP software interface component with the greatest similarity as the hierarchical matching result of the current user review subject.

[0027] Furthermore, based on the user usage feedback interaction information words, the user comment status is interactively analyzed to obtain interaction classification results, including: designing user comment status keywords; using an embedding model to embed the user comment status and user comment status keywords in the user usage feedback interaction information, respectively, to generate corresponding embedding vectors: a comment status embedding vector and a comment status keyword embedding vector; based on the comment status embedding vector and the comment status keyword embedding vector, calculating the similarity between the user comment status and the user comment status keyword; traversing each user comment status, for the current user comment status, finding the user comment status keyword with the highest similarity to it, and judging the user status keyword category with the highest similarity to the current user comment status: if it belongs to the interactive category, then the interactive category matching the current user status is interactive; otherwise, if it belongs to the non-interactive category, then the interactive category matching the current user status is non-interactive.

[0028] Furthermore, the method constructs an APP software usage feedback fusion network based on the hierarchical matching results of the user review subject and the interactive classification results of the user review status, including: defining the node categories of the fusion network, the node categories include user review subject nodes, user review status nodes, APP software interface activity nodes, and APP software interface component nodes; defining the edge categories of the fusion network, the edge categories include user review subject-user review status edges, user review subject-APP software interface activity edges, user review subject-APP software interface component edges, user review status-APP software interface component edges, and APP software interface activity-APP software interface component edges; creating the user review data after cleaning. There is a user comment subject-user comment status edge; based on the filtered user operation data, all APP software interface activities-APP software interface component edges are created; the hierarchical matching results of the user comment subject are traversed, and the hierarchical matching results of the current user comment subject are judged: if the hierarchical matching result of the current user comment subject is an APP software interface activity node, then a user comment subject-APP software interface activity edge is constructed; then, based on the user comment subject-user comment status edge, the interactive classification results of the user comment status corresponding to the hierarchical matching result of the current user comment subject are traversed to determine the user comment status-APP software interface component edge; otherwise, a user comment subject-APP software interface component edge is constructed.

[0029] Furthermore, the interactive classification results of the user comment status corresponding to the hierarchical matching results of the current user comment subject are traversed to determine the user comment status-APP software interface component edge, including: based on the user comment subject-user comment status edge, the interactive classification results of the user comment status corresponding to the hierarchical matching results of the current user comment subject are traversed, and for the interactive classification results of the current user comment status, candidate user comment status-APP software interface component edges are found based on the transmission process of the user comment subject-user comment status edge, the user comment subject-APP software interface activity edge and the APP software interface activity-APP software interface component edge; if the interactive classification result of the current user comment status is an interactive category, the user comment status-APP software interface component edges of non-interactive categories are filtered out from the candidate user comment status-APP software interface component edges according to the APP software interface component classification in the user usage feedback interaction information; otherwise, if it is a non-interactive category, the user comment status-APP software interface component edges of interactive categories are filtered out from the candidate user comment status-APP software interface component edges according to the APP software interface component classification.

[0030] According to the second aspect of the embodiment of the present invention, there is provided an APP software usage feedback data fusion device based on hierarchical matching and interactive analysis, including a module of any one of the above-mentioned APP software usage feedback data fusion methods based on hierarchical matching and interactive analysis. Specifically comprising: a preprocessing module for preprocessing APP software usage feedback data; a first obtaining module for obtaining user usage feedback hierarchical information and user usage feedback interactive information based on the preprocessed APP software usage feedback data; a second obtaining module for performing hierarchical matching on the user review subject based on the user usage feedback hierarchical information to obtain a hierarchical matching result; a module for interactively analyzing the user review status based on the user usage feedback interactive information to obtain an interactive classification result; and a construction module for constructing an APP software usage feedback fusion network based on the hierarchical matching result of the user review subject and the interactive classification result of the user review status.

[0031] According to a third aspect of an embodiment of the present invention, a processor is provided, which is used to run a program, wherein the program, when running, executes any one of the above-mentioned APP software usage feedback data fusion methods based on hierarchical matching and interaction analysis.

[0032] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the above-mentioned APP software usage feedback data fusion methods based on hierarchical matching and interactive analysis.

[0033] Example 2:

[0034] like Figure 1-3 As shown, according to an embodiment of the present invention, a method for fusion of APP software usage feedback data based on hierarchical matching and interaction analysis is provided, which can be executed by a terminal device, that is, one or more processors in the terminal device execute the following steps Step 1 to Step 4, as follows:

[0035] Step 1: Preprocess the APP software usage feedback data;

[0036] Step 1.1. Load the APP software usage feedback data, which includes user comment data UserComment and user operation data UserOperation, and execute Step 1.2.

[0037] In the above, the user review data is the APP software review text, which reflects the user's subjective feelings about the APP software; the user operation data UserOperation includes four fields: APP software name Package, APP software interface activity Activity, APP software interface component Widget, and APP software interface component text Content, as shown in Table 1:

[0038] Table 1 User operation data format

[0039] Field meaning Example Package APP software name com.youku.phone Activity APP software interface activities LoginActivity Widget APP software interface components button Content APP software interface component text Play

[0040] Step 1.2: Clean the user comment data, remove irrelevant characters and noise data, save the cleaned user comment data CleanComment, and execute Step 1.3.

[0041] Step 1.3: Filter the user operation data, remove invalid operation records, and save the filtered user operation data (FilteredOperation). In the above, invalid operation records are long-term repeated operation records or records with empty content. Repeated operation records exceeding a preset threshold within a preset time are considered long-term repeated operation records (e.g., if there are more than 4 consecutive repeated operation records within 5 seconds, the operation is considered a long-term repeated operation). Records with empty text in the APP software interface component in the user operation data are considered empty content records.

[0042] Step 2: Obtain user feedback hierarchy information HierarchyInfo and user feedback interaction information InteractInfo based on the pre-processed APP software usage feedback data.

[0043] Step 2.1. Define the user comment subject and state, as shown in Table 2:

[0044] Table 2 User comment subject and status

[0045] Field meaning Example User review body The main evaluation object of the review barrage User review status Modify the status of the target evaluation object Unable to send

[0046] Step 2.2: Based on the prompt engineering, design the extraction prompts for the subject and status of user comments. Based on the extraction prompts for the subject and status of user comments, use the locally deployed largest language model (e.g., DeepSeek: 14b) to extract the comment subject and status from the cleaned user comment data CleanComment, and execute Step 2.3.

[0047] Exemplarily, the design of the extraction prompt of the subject and status of the user comment includes:

[0048] ###Quest Background

[0049] User review data contains users' comments and status on specific objects. Please extract the user review body and status based on the review text context and software functions.

[0050] ###Quest Requirements

[0051] 1. Please note that there may be colloquial descriptions in the comments. If necessary, you can paraphrase them.

[0052] 2. If there are missing components in the comment, if necessary, the possible comment subject and corresponding status can be inferred based on the context. If the context does not support clear inference, NULL is output.

[0053] 3. If there are multiple evaluation subjects and corresponding statuses in a comment, extract all evaluation subjects and corresponding statuses

[0054] 4. You don’t need to explain the results

[0055] ###Example 1:

[0056] Input: "QQ", "Why can't I receive messages?"

[0057] Output: [(message, no message received)]

[0058] ####Example 2:

[0059] Input: "QQ", "I can't log in for no apparent reason, and the request keeps failing."

[0060] Output: [(login, login failed)]

[0061] Step 2.3. Classify the Widget field in the filtered user operation data FilteredOperation into interactive and non-interactive categories, save it as the APP software interface component classification WidgetClass, and execute Step 2.4.

[0062] The Widget field is classified specifically as follows: based on whether the Widget field provides active interaction function to the user, if active interaction is provided, the Widget field is considered to be interactive; otherwise, it is considered to be non-interactive.

[0063] Table 3 Widget field classification

[0064]

[0065] Step 2.4: Use the Jieba word segmentation tool to segment the Content field in the filtered user operation data FilteredOperation, and extract the segmented verbs as the APP software interface component function supplementary information WidgetFunctionInfo, and execute Step 2.5.

[0066] Step 2.5: Based on the prompt engineering, design prompts for reasoning about the activity functions of the APP software interface, and use the second largest language model (such as DeepSeekR1) to generate supplementary information ActivityFunctionInfo of the APP software interface activity functions, and execute Step 2.6.

[0067] Exemplarily, the prompts for reasoning about the activity function of the APP software interface include:

[0068] ###Quest Background

[0069] You are an expert in Android application development. An Activity is an interface within a mobile app. Typically, an Activity contains multiple widgets, which provide user interaction or display content. Based on the provided app and Activity names, you are asked to infer the user's likely purpose and desired functionality within that interface.

[0070] ###Quest Requirements

[0071] You need to consider the functionality of the target software and the functionality of the given Activity, and try to give an answer consisting of simple actions or verb-object phrases. You don't need to explain, just give the final answer.

[0072] ###Example

[0073] Input: login activity

[0074] Output: Login registration account management

[0075] Considering that Step 2.2 corresponds to the text information extraction task, the present invention uses the DeepSeek:14b model with smaller parameters. The use of this model helps to improve efficiency while ensuring performance; for the functional reasoning task corresponding to Step 2.5, DeepSeek R1 has better reasoning effect.

[0076] Step 2.6. Use the user comment subject Subject obtained in Step 2.2, the APP software interface component function supplementary information WidgetFunctionInfo obtained in Step 2.4, and the APP software interface activity function supplementary information ActivityFunctionInfo obtained in Step 2.5 as the user usage feedback hierarchy information HierarchyInfo; use the user comment status State and the APP software interface component classification WidgetClass obtained in Step 2.2 as the user usage feedback interaction information InteractInfo.

[0077] Step 3: Based on the hierarchical information of user feedback, perform hierarchical matching on the user comment subjects to obtain hierarchical matching results; based on the user feedback interaction information and the designed user comment status keywords, perform interactive analysis on the user comment status to obtain interactive classification results.

[0078] Step 3.1. Use the open source embedding model to embed the Subject, WidgetFunctionInfo, and ActivityFunctionInfo in the user feedback hierarchy information HierarchyInfo, and generate the corresponding embedding vectors: comment subject embedding vector VecSubject, APP software interface component function supplementary information embedding vector VecWidgetFunctionInfo, APP software interface activity function supplementary information embedding vector VecActivityFunctionInfo, and execute Step 3.2.

[0079] Exemplarily, the embedding model is bge-large-zh-v1.5.

[0080] Step 3.2, based on the comment subject embedding vector VecSubject and the APP software interface activity function supplementary information embedding vector VecActivityFunctionInfo, calculate the similarity SimSubjectActivity between the user comment subject and the APP software interface activity function supplementary information, using the cosine similarity calculation method, as shown in formula (1).

[0081]

[0082] Step 3.3, based on the comment subject embedding vector VecSubject and the APP software interface component function supplementary information embedding vector VecWidgetFunctionInfo, calculate the similarity SimSubjectWidget between the user comment subject and the APP software interface component function supplementary information, using the cosine similarity calculation method, as shown in formula (2).

[0083]

[0084] Step 3.4, traverse each user comment subject Subject, for the current user comment subject Subject_i, find its maximum similarity SimSubjectActivityMax and SimSubjecWidgetMax with the APP software interface activity and APP software interface component, and execute Step 3.5.

[0085] Step 3.5. Compare SimSubjectActivityMax and SimSubjecWidgetMax. If SimSubjectActivityMax is greater than SimSubjecWidgetMax, execute Step 3.6; otherwise, execute Step 3.7.

[0086] Step 3.6: Match the current user comment subject Subject_i with the APP software interface activity with the greatest similarity as the hierarchical matching result of the current user comment subject Subject_i, and execute Step 3.8.

[0087] Step 3.7: Match the current user comment subject Subject_i with the APP software interface component with the greatest similarity as the hierarchical matching result of the current user comment subject Subject_i, and execute Step 3.8.

[0088] Step 3.8. Return to Step 3.4 until the loop ends, save the hierarchical matching results of all user comment subjects (SubjectHierarchyResult), and execute Step 3.9.

[0089] Step 3.9, design the user comment status keyword StateKeyWord, as shown in Table 4:

[0090] Table 4 User comment status keywords

[0091]

[0092] Step 3.10: Use the open source embedding model to embed the State in the user feedback interaction information InteractInfo and the user comment status keyword StateKeyWord, generate the corresponding embedding vectors: comment status embedding vector VecState and comment status keyword embedding vector VecStateKeyWord, and execute Step 3.11.

[0093] Exemplarily, the embedding model is bge-large-zh-v1.5.

[0094] Step 3.11. Based on the comment status embedding vector VecState and the comment status keyword embedding vector VecStateKeyWord, calculate the similarity SimStateKeyWord between the user comment status and the user comment status keyword using the cosine similarity calculation method, as shown in formula (3):

[0095]

[0096] Step 3.12: Traverse each user review state State, find the user review state keyword with the highest similarity to the current user review state State_i, and execute Step 3.13.

[0097] Step 3.13: Determine the user state keyword category with the highest similarity to State_i. If it belongs to the interactive category, execute Step 3.14; if it belongs to the non-interactive category, execute Step 3.15.

[0098] Step 3.14: Match the interaction category of the current user state State_i to interaction, and execute Step 3.16.

[0099] Step 3.15: If the interaction category of the current user state State_i is non-interactive, execute Step 3.16.

[0100] Step 3.16. Return to Step 3.12 until the traversal is completed, save the interaction classification results StateInteractResult of all user comment states, and end Step 3.

[0101] Step 4: APP software is constructed using feedback fusion network

[0102] Step 4.1. Define the node types of the fusion network. These NodeTypes include user comment subject nodes (SubjectNodes), user comment state nodes (StateNodes), app software interface activity nodes (ActivityNodes), and app software interface component nodes (WidgetNodes). Then proceed to Step 4.2. For illustrative purposes, a user comment subject node represents a node of the user comment subject type, a user comment state node represents a node of the user comment state type, and so on.

[0103] Step 4.2. Define the edge category EdgeType of the fusion network. The edge category EdgeType includes user comment subject-user comment status edge CommentEdge, user comment subject-APP software interface activity edge SubjectActivityEdge, user comment subject-APP software interface component edge SubjectWidgetEdge, user comment status-APP software interface component edge StateWidgetEdge, and APP software interface activity-APP software interface component edge ActivityWidgetEdge. Execute Step 4.3.

[0104] Step 4.3: Based on the cleaned user comment data CleanComment, create all user comment bodies and user comment status edges CommentEdge and execute Step 4.4.

[0105] Step 4.4: Based on the filtered user operation data (FilteredOperation), create ActivityWidgetEdge edges between all app interface activities and app interface components, and proceed to Step 4.5. For example, in the example in Table 1, LoginActivity and button are one edge.

[0106] Step 4.5: Traverse the hierarchical matching results SubjectHierarchyResult of the user comment subject, and execute Step 4.6 for the hierarchical matching result SubjectHierarchyResult_i of the current user comment subject.

[0107] Step 4.6. If the hierarchical matching result of SubjectHierarchyResult_i is the APP software interface activity node ActivityNode, then construct a user comment subject-APP software interface activity edge SubjectActivityEdge_i and execute Step 4.7. Otherwise, execute Step 4.12.

[0108] Step 4.7. Based on the user comment subject-user comment status edge, traverse the interaction classification result StateInteractResult of the user comment status corresponding to SubjectHierarchyResult_i, and execute Step 4.8 for the interaction classification result StateInteractResult_j of the current user comment status.

[0109] Step 4.8. According to the transmission process of user comment subject-user comment status edge CommentEdge, user comment subject-APP software interface activity edge SubjectActivityEdge and APP software interface activity-APP software interface component edge ActivityWidgetEdge, find the candidate user comment status-APP software interface component edge CandidateStateWidgetEdge and execute Step 4.9.

[0110] Step 4.9. If the interaction classification result of StateInteractResult_j is the interaction category, filter out the non-interaction category user comment status-APP software interface component edge from the user comment status-APP software interface component edge CandidateStateWidgetEdge according to the APP software interface component classification WidgetClass (that is, retain the interaction category user comment status-APP software interface component edge), and execute Step 4.11, otherwise execute Step 4.10.

[0111] Step 4.10. Filter out the interactive user comment status-APP software interface component edge from the CandidateStateWidgetEdge based on the APP software interface component classification WidgetClass, and execute Step 4.11.

[0112] Step 4.11. Return to Step 4.7 until the traversal is completed and execute Step 4.5.

[0113] Step 4.12. Build a user comment subject-APP software interface component edge SubjectActivityEdge_i and execute Step 4.13.

[0114] Step 4.13. Return to Step 4.5 until the traversal is completed, completing the construction of the feedback fusion network used by the APP software.

[0115] The specific embodiments of the present invention are described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Various changes can be made within the knowledge of ordinary technicians in this field without departing from the scope of the present invention.

Claims

1. A method for fusion of APP software usage feedback data based on hierarchical matching and interaction analysis, characterized in that: The following steps are involved: Pre-process the APP software usage feedback data; Based on the pre-processed APP software usage feedback data, obtain user usage feedback level information and user usage feedback interaction information; Based on the user feedback hierarchical information, hierarchical matching is performed on the user comment subject to obtain a hierarchical matching result; based on the user feedback interaction information, interactive analysis is performed on the user comment status to obtain an interactive classification result; Based on the hierarchical matching results of user review subjects and the interactive classification results of user review status, an APP software usage feedback fusion network is constructed.

2. The APP software usage feedback data fusion method based on hierarchical matching and interactive analysis according to claim 1 is characterized in that: The APP software uses feedback data including user comment data and user operation data, cleans the user comment data to obtain cleaned user comment data; and filters the user operation data to obtain filtered user operation data.

3. The APP software usage feedback data fusion method based on hierarchical matching and interactive analysis according to claim 1 is characterized in that: The user usage feedback level information and user usage feedback interaction information are obtained based on the pre-processed APP software usage feedback data, including: Extracting the user comment subject and the user comment status from the cleaned user comment data in the preprocessed APP software usage feedback data based on the first language model; Classifying the "APP software interface component" field in the filtered user operation data in the preprocessed APP software usage feedback data to obtain an APP software interface component classification; Use a word segmentation tool to segment the "APP software interface component text" field in the filtered user operation data, and extract the segmented verbs as supplementary information about the APP software interface component functions; Use the second language model to infer the APP software interface activities in the filtered user operation data to generate supplementary information about the APP software interface activities; The obtained user comment body, APP software interface component function supplementary information, and APP software interface activity function supplementary information are used as user usage feedback hierarchical information; the obtained user comment status and APP software interface component classification are used as user usage feedback interaction information.

4. The APP software usage feedback data fusion method based on hierarchical matching and interactive analysis according to claim 1 is characterized in that: Based on the user feedback hierarchical information, hierarchical matching is performed on the user comment subject to obtain a hierarchical matching result, including: Use the embedding model to embed the hierarchical information of user feedback and generate corresponding embedding vectors: the embedding vector of the comment body, the embedding vector of the supplementary information of the APP software interface component function, and the embedding vector of the supplementary information of the APP software interface activity function; Based on the embedding vector of the comment subject and the embedding vector of the APP software interface activity function supplementary information, calculate the similarity between the user comment subject and the APP software interface activity function supplementary information; Based on the embedding vector of the comment subject and the embedding vector of the supplementary information of the APP software interface component function, the similarity between the user comment subject and the supplementary information of the APP software interface component function is calculated; Traverse each user review subject, and for the current user review subject, find its maximum similarity SimSubjectActivityMax and SimSubjecWidgetMax with the APP software interface activity and APP software interface component; compare SimSubjectActivityMax and SimSubjecWidgetMax: if SimSubjectActivityMax is greater than SimSubjecWidgetMax, match the current user review subject with the APP software interface activity with the greatest similarity as the hierarchical matching result of the current user review subject; otherwise, match the current user review subject with the APP software interface component with the greatest similarity as the hierarchical matching result of the current user review subject.

5. The APP software usage feedback data fusion method based on hierarchical matching and interactive analysis according to claim 1 is characterized in that: Based on the user feedback interaction information words, the user comment status is interactively analyzed to obtain an interaction classification result, including: Design user review status keywords; Use the embedding model to embed the user comment status and user comment status keywords in the user feedback interaction information, and generate corresponding embedding vectors: comment status embedding vector and comment status keyword embedding vector; Based on the comment status embedding vector and the comment status keyword embedding vector, calculate the similarity between the user comment status and the user comment status keyword; Traverse each user comment status, find the user comment status keyword with the highest similarity to the current user comment status, and determine the user status keyword category with the highest similarity to the current user comment status: if it belongs to the interactive category, then the interaction category that matches the current user status is interactive; otherwise, if it belongs to the non-interactive category, then the interaction category that matches the current user status is non-interactive.

6. The APP software usage feedback data fusion method based on hierarchical matching and interactive analysis according to claim 1 is characterized in that: The method of constructing an APP software usage feedback fusion network based on the hierarchical matching results of the user review subjects and the interactive classification results of the user review status includes: Defining node categories of the fusion network, the node categories include user comment subject nodes, user comment status nodes, APP software interface activity nodes, and APP software interface component nodes; Defining edge categories of the fusion network, the edge categories including user comment subject-user comment status edge, user comment subject-APP software interface activity edge, user comment subject-APP software interface component edge, user comment status-APP software interface component edge, and APP software interface activity-APP software interface component edge; Based on the cleaned user review data, create all user review subject-user review status edges; Based on the filtered user operation data, create all APP software interface activities-APP software interface component edges; Traverse the hierarchical matching results of the user comment body and judge the hierarchical matching results of the current user comment body: If the hierarchical matching result of the current user review subject is an APP software interface activity node, a user review subject-APP software interface activity edge is constructed. Then, based on the user review subject-user review status edge, the interactive classification results of the user review status corresponding to the hierarchical matching result of the current user review subject are traversed to determine the user review status-APP software interface component edge. Otherwise, construct a user comment subject-APP software interface component edge.

7. The APP software usage feedback data fusion method based on hierarchical matching and interactive analysis according to claim 6 is characterized in that: The step of traversing the interactive classification results of the user comment status corresponding to the hierarchical matching results of the current user comment subject and determining the user comment status-APP software interface component edge includes: Based on the user review subject-user review status edge, traverse the interactive classification results of the user review status corresponding to the hierarchical matching result of the current user review subject. For the interactive classification results of the current user review status, find the candidate user review status-APP software interface component edge based on the transmission process of the user review subject-user review status edge, the user review subject-APP software interface activity edge, and the APP software interface activity-APP software interface component edge; If the interaction classification result of the current user comment status is the interactive category, the user comment status-APP software interface component edges of the non-interactive category are filtered out from the candidate user comment status-APP software interface component edges according to the APP software interface component classification in the user usage feedback interaction information; otherwise, if it is the non-interactive category, the user comment status-APP software interface component edges of the interactive category are filtered out from the candidate user comment status-APP software interface component edges according to the APP software interface component classification.

8. An APP software usage feedback data fusion device based on hierarchical matching and interactive analysis, characterized in that: A module comprising the method for fusion of feedback data used by APP software based on hierarchical matching and interactive analysis as described in any one of claims 1 to 7.

9. A processor, characterized in that: The processor is used to run a program, wherein the program, when running, executes the APP software usage feedback data fusion method based on hierarchical matching and interactive analysis as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute the APP software usage feedback data fusion method based on hierarchical matching and interaction analysis as described in any one of claims 1 to 7.

Images