Note-based user feature analysis method

Abstract

The invention relates to a note-based user feature analysis method and device, computer equipment and a storage medium, and relates to the technical field of computers. The method comprises the following steps: acquiring local note data of a user by local terminal equipment; constructing an initial feature space and an evaluation index corresponding to the target dimension based on note information corresponding to the target dimension extracted from the local note data; the initial feature space comprises each note feature corresponding to a target dimension; performing linear regression model training based on the initial feature space and the evaluation index to obtain a local model corresponding to the target dimension; performing user feature analysis based on the local model of the target dimension to obtain an analysis result; the analysis result is used for indicating the note use tendency of the user in the target dimension; through the method, the user characteristics can be analyzed locally based on the existing notes, interference of invalid information is reduced, and therefore the accuracy of user characteristic analysis is improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of computer technology, and in particular to a user feature analysis method based on notes. Background Art

[0002] User characteristics are descriptions of the user's basic attributes, information needs, information behavior, psychological state, environmental influences, etc. User characteristics can be used to build user portraits and identify commonalities among user groups in order to adopt targeted operations or development strategies. For example, users of a certain video platform have a high click-through rate for videos in the food category. When the platform invests in content support, it can focus on food videos to maximize the input-output ratio.

[0003] In the prior art, user feature analysis is achieved by recording user behavior operations, collecting key data and reporting it to the server, and then analyzing and generating user preference tendencies. However, when this user feature analysis method is applied to note users, due to the large amount of note editing operations, a large amount of server resources are consumed for storage and processing; and the note application involves user personal data that can be deleted and frequently changed, which may result in a large number of erroneous operations and invalid information being recorded, affecting the final user feature results; and because notes can be copied and moved and the content has a large amount of information, this method cannot perform content preference analysis on existing notes, resulting in unclear and biased user portraits; thus, the existing user feature analysis methods are less effective in performing user feature analysis based on notes. Summary of the invention

[0004] An embodiment of the present application provides a user feature analysis method based on notes, which can implement user feature analysis in a target dimension based on local note data in a local terminal device. It has good universality in scenarios where note data is frequently deleted and modified, can reduce interference from erroneous operations and invalid information, and improve the accuracy of user feature analysis. The technical solution is as follows.

[0005] On the one hand, a user feature analysis method based on notes is provided, the method being executed by a local terminal device, the method comprising:

[0006] Get the user's local note data;

[0007] Based on the note information corresponding to the target dimension extracted from the local note data, construct an initial feature space and an evaluation index corresponding to the target dimension; the initial feature space contains each note feature corresponding to the target dimension;

[0008] Performing linear regression model training based on the initial feature space and the evaluation index to obtain a local model corresponding to the target dimension;

[0009] Perform user feature analysis based on the local model of the target dimension to obtain analysis results; the analysis results are used to indicate the user's note usage tendency in the target dimension.

[0010] On the other hand, a user feature analysis device based on notes is provided, which is applied to a local terminal device, and the device includes:

[0011] The acquisition module is used to obtain the user's local note data;

[0012] A construction module, configured to construct an initial feature space and an evaluation index corresponding to the target dimension based on the note information corresponding to the target dimension extracted from the local note data; the initial feature space includes each note feature corresponding to the target dimension;

[0013] A training module, used for performing linear regression model training based on the initial feature space and the evaluation index to obtain a local model corresponding to the target dimension;

[0014] The analysis module is used to perform user feature analysis based on the local model of the target dimension to obtain analysis results; the analysis results are used to indicate the user's note usage tendency in the target dimension.

[0015] In a possible implementation, the training module includes:

[0016] A dimension-increasing submodule, used for performing feature dimension-increasing processing on each note feature in the initial feature space to obtain an extended feature space;

[0017] The training submodule is used to perform linear regression model training based on each note feature in the extended feature space and the evaluation index to obtain a local model corresponding to the target dimension.

[0018] In a possible implementation, the building block includes:

[0019] An information table construction submodule, for constructing an information table corresponding to the target dimension based on the note information corresponding to the target dimension extracted from the local note data; the information in the information table is updated in real time as the local note data changes;

[0020] The information determination submodule is used to determine the initial feature space and evaluation index corresponding to the target dimension based on the information table corresponding to the target dimension.

[0021] In a possible implementation, the target dimension includes a classification dimension, a content dimension, and an editing dimension;

[0022] The local models correspondingly include classification local models, content local models and editing local models.

[0023] In one possible implementation,

[0024] In a possible implementation, the information table construction submodule is used to:

[0025] In the case where the target dimension is a classification dimension, a note information table is constructed based on classification information corresponding to the classification dimension extracted from the local note data, wherein a header of the note information table includes a note identifier, a note tag array, and a note viewing count;

[0026] Based on the note information table, classification information is accumulated and counted to obtain a classification information table; the header of the classification information table includes the number of notes, the number of tags, and the total number of note views; wherein the number of tags is used to indicate the number of non-repeated tags between multiple notes;

[0027] The information determination submodule is used to:

[0028] Determine the number of notes and the number of tags in the note information table as note features in the initial feature space corresponding to the classification dimension;

[0029] The total note viewing times in the note information table is determined as an evaluation index corresponding to the classification dimension.

[0030] In a possible implementation, the information table construction submodule is used to:

[0031] In the case where the target dimension is a content dimension, a content information table is constructed based on the content information corresponding to the content dimension extracted from the local note data; the header of the content information table includes a note identifier, a note content score and the content information; the content information includes at least one of the following: the number of note views, the length of the text, the number of pictures, the number of audios, the number of videos, the number of hyperlinks, and the number of files; the note content score is determined based on the content information and is used to indicate the probability of a user adding a note corresponding to the content information;

[0032] The information determination submodule is used to:

[0033] Determining the content information in the content information table as note features in the initial feature space corresponding to the content dimension;

[0034] An evaluation index corresponding to the content dimension is determined based on the note content score in the content information table.

[0035] In a possible implementation, the information table construction submodule is used to:

[0036] In the case where the target dimension is an editing dimension, an editing information table is constructed based on the editing information corresponding to the editing dimension extracted from the local note data, the header of the editing information table including a note identifier, a note editing score and the editing information; the editing information includes at least one of the following: number of edits, font size, font color, emphasis style, typesetting style, list style and table style; the note editing score is determined based on the editing information and is used to indicate the probability of a user adding a note with corresponding editing information;

[0037] The information determination submodule is used to determine the editing information in the editing information table as a note feature in the initial feature space corresponding to the editing dimension;

[0038] An evaluation index corresponding to the editing dimension is determined based on the note editing score in the editing information table.

[0039] In a possible implementation manner, the device further includes:

[0040] The model update module is used to perform the process of note feature extraction and linear regression model training based on the updated classification information table when the model training node is reached and the target dimension is the classification dimension, so as to obtain the updated classification local model.

[0041] In a possible implementation, the model updating module is used to:

[0042] When the model training node is reached and the target dimension is the content dimension or the editing dimension, the process of note feature extraction and linear regression model training is performed based on the updated information in the updated content information table or the editing information table to obtain the updated content local model or the editing local model.

[0043] In a possible implementation, the analysis module is used to:

[0044] Performing user feature analysis based on model parameters in the local model of the target dimension to obtain the analysis result;

[0045] or,

[0046] A corresponding two-dimensional image is generated based on the local model of the target dimension, so as to perform user feature analysis based on the two-dimensional image to obtain the analysis result.

[0047] On the other hand, a computer device is provided, which includes a processor and a memory, wherein the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor to implement the above-mentioned note-based user feature analysis method.

[0048] On the other hand, a computer-readable storage medium is provided, in which at least one computer program is stored, and the computer program is loaded and executed by a processor to implement the above-mentioned note-based user feature analysis method.

[0049] On the other hand, a computer program product is provided, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions, which, when executed by a computer, cause the computer to execute to implement the note-based user feature analysis method provided in the various optional implementations described above.

[0050] The technical solution provided by this application may have the following beneficial effects:

[0051] In the user feature analysis method based on notes provided in the embodiment of the present application, the local terminal device obtains the local note data of the user, extracts the note information of the target dimension from the local note data, constructs the corresponding initial feature space based on the extracted note information and obtains the corresponding evaluation index, and then trains the linear regression model based on the initial feature space and the evaluation index of the local note data to obtain the local model under the target dimension, and performs user feature analysis based on the local model to obtain the analysis result indicating the user's note usage tendency under the target dimension. Through the above method, user feature analysis under the target dimension based on local note data can be implemented in the local terminal device, which has good universality in the scenario of frequently deleted and modified note data, can reduce the interference of erroneous operations and invalid information, and thus improve the accuracy of user feature analysis; in addition, by performing user feature analysis on the local terminal device, the privacy protection effect of user data can be improved, and the security of user data can be improved.

[0052] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0054] Figure 1 A flowchart of a user feature analysis method based on notes provided by an exemplary embodiment of the present application is shown;

[0055] Figure 2 A flowchart of a user feature analysis method based on notes provided by an exemplary embodiment of the present application is shown;

[0056] Figure 3 A block diagram of a user feature analysis device based on notes provided by an exemplary embodiment of the present application is shown;

[0057] Figure 4 A structural block diagram of a computer device shown in an exemplary embodiment of the present application is shown. DETAILED DESCRIPTION

[0058] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0059] The embodiment of the present application provides a user feature analysis method based on notes, which can generate user features by analyzing the final product of the user note application. The final product is each local user note (i.e., existing notes) saved locally by the user, thereby achieving the purpose of performing user feature analysis based on local user notes. This user feature analysis method based on local user notes can solve the problem of large process operations that cannot be recorded and changes in content that can be deleted and modified, thereby improving the analysis effect of user features.

[0060] Figure 1 A flowchart of a user feature analysis method based on notes provided by an exemplary embodiment of the present application is shown, and the method can be executed by a local terminal device; Figure 1 As shown, the user feature analysis method based on notes may include the following steps:

[0061] Step 110, obtaining the user's local note data.

[0062] The local terminal device can obtain the user's local note data from the note application of the local terminal device; the local note data can be note data generated by a note application with a note function, and the implementation form of the note content can be text files, notes, and notepads, etc. This application does not limit the implementation form of the note content.

[0063] The local note data may include note information corresponding to multiple notes; since user characteristics change over time, in order to ensure the timeliness of user characteristic analysis based on notes, in one possible implementation, when the local terminal device obtains the user's local note data, it can obtain local note data within a target time period from the current time point, such as local note data within one month, or local note data within three months, etc. The length of the target time period can be set based on actual needs, and this application does not impose any restrictions on this.

[0064] Step 120, based on the note information corresponding to the target dimension extracted from the local note data, construct an initial feature space and evaluation index corresponding to the target dimension; the initial feature space contains various note features corresponding to the target dimension.

[0065] Based on the usage scenarios of the note-taking application, user feature analysis can be performed in different dimensions. The target dimension is one of multiple dimensions. Based on the different analysis dimensions, different note information can be extracted from the local note data. In principle, if the target dimension is the practicality of the notes, then note information related to the practicality of the notes can be extracted based on the local note data, for example, note content, time information (such as creation time, modification time, and modification time interval, etc.), structure information (such as note length, number of paragraphs, etc.), behavior information (such as number of views, tags), etc.

[0066] Based on the extracted note information, an initial feature space corresponding to the target dimension can be constructed. In one possible implementation, each note feature in the initial feature space corresponds to the note information; in another possible implementation, each note feature in the initial feature space is generated based on the note information; each analysis dimension has a corresponding evaluation indicator, which can be one of the note information obtained in the note information extraction, such as the number of views; or, the evaluation indicator can also be an evaluation score calculated based on the extracted note information.

[0067] Step 130, training a linear regression model based on the initial feature space and the evaluation index to obtain a local model corresponding to the target dimension.

[0068] In one possible implementation, the local terminal device may use the initial feature space of the local note data as input parameters (X) and the evaluation index as the output result (Y) to perform linear regression model training to obtain a local model corresponding to the corresponding dimension; or, in another possible implementation, the local terminal device may further perform input parameter dimension upgrading based on the initial feature space of each local note data to further obtain the interaction features between the various note features, and perform linear regression model training based on the upgraded input parameters and the evaluation results to obtain a local model corresponding to the corresponding dimension; the local model may indicate the relationship between the note features and the evaluation index under the corresponding dimension.

[0069] In one possible implementation, the local terminal device can perform linear regression model training through a model training tool, and the model training tool can be sklearn's StandardScaler, LinearRegression tool, or it can also be other tools for linear regression fitting, which is not limited in this application.

[0070] Step 140, performing user feature analysis based on the local model of the target dimension to obtain analysis results; the analysis results are used to indicate the user's note usage tendency in the target dimension.

[0071] After obtaining the local model of the target dimension, the local terminal device can export the parameters of the local model as user features corresponding to the target dimension; based on the analysis of the user features of the target dimension, the user's note usage tendency under the target dimension can be obtained.

[0072] Furthermore, different local terminal devices can obtain local models corresponding to different users based on the above-mentioned user feature analysis method of notes. By performing an overall analysis of the user characteristics of the local models of different users, the common characteristics of the user group can be obtained, such as whether the user frequently uses a certain function of the note application, what type of note content the user prefers to edit, etc.

[0073] In summary, the note-based user feature analysis method provided in the embodiment of the present application is that the local terminal device obtains the user's local note data, extracts the note information of the target dimension from the local note data, and constructs the corresponding initial feature space based on the extracted note information and obtains the corresponding evaluation index. After that, a linear regression model is trained based on the initial feature space and the evaluation index of the local note data to obtain a local model under the target dimension, and user feature analysis is performed based on the local model to obtain an analysis result indicating the user's note usage tendency under the target dimension. Through the above method, user feature analysis under the target dimension based on local note data can be implemented in the local terminal device, which has good universality in the scenario of frequently deleted and modified note data, can reduce interference from erroneous operations and invalid information, and thus improve the accuracy of user feature analysis; in addition, by performing user feature analysis on the local terminal device, the privacy protection effect of user data can be improved, and the security of user data can be improved.

[0074] Figure 2 A flowchart of a user feature analysis method based on notes provided by an exemplary embodiment of the present application is shown. The method can be executed by a local terminal device, such as Figure 2 As shown, the method comprises the following steps:

[0075] Step 210, obtaining the user's local note data.

[0076] Step 220, based on the note information corresponding to the target dimension extracted from the local note data, construct an initial feature space and evaluation index corresponding to the target dimension; the initial feature space contains various note features corresponding to the target dimension.

[0077] In a possible implementation, the local terminal device may store the note information corresponding to the target dimension extracted from the local note data by creating a corresponding information table; the process may be implemented as follows:

[0078] Based on the note information corresponding to the target dimension extracted from the local note data, an information table corresponding to the target dimension is constructed; the information in the information table is updated in real time as the local note data changes;

[0079] Based on the information table corresponding to the target dimension, the initial feature space and evaluation index corresponding to the target dimension are determined.

[0080] The header of the information table includes a note identifier, note information, and evaluation indicators; wherein each note identifier corresponds to a note in the local note data, and each note corresponds to corresponding parameters under the note information and corresponding evaluation indicator parameters. Schematically, Table 1 shows a schematic diagram of an information table under a certain analysis dimension. Assuming that there are 4 notes recorded in the local note data, as shown in Table 1, the information table records the note information parameters of Notes 1-4 and their corresponding evaluation indicator parameters. It should be noted that Table 1 is only for illustration, and the type and quantity of note information to be extracted under different analysis dimensions are different, and this application does not limit this.

[0081] Table 1

[0082] Notes logo Note information Evaluation indicators 1 2 3 4

[0083] The information in the information table is updated in real time as the local note data changes. For example, if the note information is the note font, and the user adjusts the font in Note 1 from Kaiti to Songti, then in the information table, the note font corresponding to Note 1 is changed to Songti; further, after the update is completed, the corresponding position in the information table can be marked as updated, and optionally, the update time can also be marked accordingly.

[0084] In one possible implementation, after obtaining the information table corresponding to the target dimension, the local terminal device can directly construct the initial feature space and evaluation indicators corresponding to the target dimension based on the information in the information table of the target dimension; in another possible implementation, after obtaining the information table corresponding to the target dimension, the local terminal device needs to further perform information statistics and integration based on the information table to obtain a new information table, and construct the initial feature space and evaluation indicators corresponding to the target dimension based on the information in the new information table.

[0085] Step 230, perform feature dimension upgrading processing on each note feature in the initial feature space to obtain an extended feature space.

[0086] Among them, feature dimensionality increase refers to increasing the number or complexity of features to capture more information or to improve the performance of the model obtained by subsequent training; schematically, when the feature dimensionality increase is processed, the local terminal device can perform dimensionality increase by polynomial regression. In this case, the local terminal device can achieve feature dimensionality increase through a feature dimensionality increase tool. The feature dimensionality increase tool can be executed by the PolynomialFeatures tool of sklearn, and the maximum degree of the polynomial feature is controlled by setting the value of degree. Alternatively, it can be other tools for feature dimensionality increase, such as implementing feature dimensionality increase through custom functions, etc.; schematically, if the note features in the initial feature space are x1, x2, the expanded feature space obtained after the feature dimensionality increase process contains x1, x2, x1 2 , x2 2 And x1*x2; alternatively, the local computer device may also create new features by means of feature combination, for example, combining the text features and time features of the notes to create the feature of "the number of notes at a specific time", etc. Alternatively, the local computer device may also use other forms of feature dimensionality upgrading to perform processing to obtain an expanded feature space; in the embodiment of the present application, dimensionality upgrading is illustrated by means of polynomial regression.

[0087] Step 240, a linear regression model is trained based on each note feature and evaluation index in the extended feature space to obtain a local model corresponding to the target dimension.

[0088] In a possible implementation, the target dimension includes a classification dimension, a content dimension, and an editing dimension, and correspondingly, the local model includes a classification local model, a content local model, and an editing local model. Among them, the classification dimension can indicate the user's tendency to organize and classify existing notes, and the impact of classification on the use of notes; the content dimension can indicate the content richness of the notes, the tendency of the content form, and the impact of the content on the viewing of notes; the editing dimension can indicate the degree of use of various rich text format settings, and the tendency of using editing functions.

[0089] The following embodiments respectively illustrate the local model generation process corresponding to the above three analysis dimensions.

[0090] 1. Classification Dimensions:

[0091] The classification dimension refers to the summary of the final result of the user's classification and management of notes by organizing notebooks / folders, labeling notes, etc. Among them, notebooks / folders and tags are two ways to classify notes. In the embodiment of the present application, the above classification methods are homogenized and are all uniformly regarded as tags.

[0092] The classification local model generated under the classification dimension can measure the user's sorting tendency and viewing tendency in the classification function; the relationship between the sorting tendency and the viewing tendency can also be quantitatively analyzed from the classification local model.

[0093] The process of generating a classification local model by a local terminal device can be implemented as follows:

[0094] Based on the classification information corresponding to the classification dimension extracted from the local note data, a note information table is constructed, wherein the header of the note information table includes a note identifier, a note tag array, and a note viewing count;

[0095] Based on the note information table, the classification information is accumulated and counted to obtain a classification information table; the header of the classification information table includes the number of notes, the number of tags, and the total number of note views; wherein the number of tags is used to indicate the number of non-repeated tags between multiple notes;

[0096] Determine the number of notes and the number of tags in the note information table as note features in the initial feature space corresponding to the classification dimension;

[0097] The total number of note views in the note information table is determined as an evaluation indicator corresponding to the classification dimension.

[0098] Illustratively, the classification information corresponding to the classification dimension extracted from the local note data may be the note tags and note viewing counts of each note. The extracted information is randomly filled into a table in units of notes to obtain a note information table; Table 2 shows a note information table provided by an exemplary embodiment of the present application. As shown in Table 2, the note information table contains 3 columns of data, and the corresponding table headers are note identification, note tag array, and note viewing count; the note tag array is used to record each note tag of the corresponding note.

[0099] Table 2

[0100] Note ID Note tag array Note Views 264 Work, meeting minutes 7 17 Life 13 136 Work, important, urgent 3 ... ... ...

[0101] Construct an empty classification information table with 3 columns of data: the number of notes, the number of tags, and the total number of views. When accumulating statistics on classification information, the number of notes refers to the number of data rows extracted from the note information table. For example, if the first n rows of data are extracted, the number of notes is n; the number of tags refers to the number of non-repeated tags after merging all note tag arrays in the first n rows of data; the total number of views indicates the cumulative result of the number of views of all notes in the first n rows of data; fill in the above cumulative statistical results into the nth row of the classification information table; the value of n ranges from 1 to the maximum number of rows in the note information table, and repeat the above operation until the maximum number of rows in the classification data table is the same as the maximum number of rows in the note information table; Table 3 is a classification information table provided in an exemplary embodiment of the present application; as shown in Table 3, the header of the classification information table Contains the number of notes, the number of tags, and the total number of note views; after cumulative statistics based on the data shown in Table 2, when the number of notes is 1, that is, after counting the first row of data in the note information table, the number of tags is 2, the total number of views is 7, and the corresponding data is filled in the first row of the classification information table; when the number of notes is 2, that is, after counting the first two rows of data in the note information table, the number of tags is 3, the total number of views is 20, and the corresponding data is filled in the second row of the classification information table; when the number of notes is 3, that is, after counting the first three rows of data in the note information table, the number of tags is 5, the total number of views is 23, and the corresponding data is filled in the third row of the classification information table, and so on.

[0102] Table 3

[0103] Number of notes (n) Number of tags Total Views 1 2 7 2 3 20 3 5 23 ... ... ...

[0104] After obtaining the classification information table, the number of notes and the number of tags in the classification information table are used as note features in the initial feature space; that is, the {number of notes, number of tags} in the classification data table are used as input parameters; further, the input parameters can be upgraded in dimension by polynomial regression to obtain an extended feature space. For example, by using the PolynomialFeatures tool of sklearn to upgrade the feature dimension, and degree = 2, the extended feature space under the classification dimension can be expressed as:

[0105] X=[[1.0,a,b,a*a,b*b,a*b]]

[0106] Among them, 1.0 is a fixed constant, a is the number of notes, and b is the number of tags.

[0107] The total number of views is used as the evaluation index corresponding to the classification dimension, that is, the total number of views is used as the output result. Based on the classification information table shown in Table 3, the output result Y=[7, 20, 23, ...].

[0108] The extended feature space X under the classification dimension and the output result Y are used as training sets to train the linear regression model to obtain a classification local model; this process can be performed through the StandardScaler and LinearRegression tools of sklearn.

[0109] The local terminal device exports the parameters of the classification local model as classification user features, and the number of parameters of the classification local model corresponds to the number of features in the extended feature space under the classification dimension; illustratively, the number of parameters of the classification local model corresponding to the classification information table shown in Table 3 is 6, which is [0, 1.51828215, 1.66045176, -0.03191915, -0.02536972, 0.08237879].

[0110] 2. Content Dimension:

[0111] The content dimension refers to the summary of the final results of the user's note text, pictures, attachments and other content with specific meanings. The content local model generated under the content dimension can measure the user's content preference (which content they prefer) and richness preference (the content richness of the preferred notes) in the content dimension.

[0112] The process of generating a local content model by a local terminal device can be implemented as follows:

[0113] Based on the content information corresponding to the content dimension extracted from the local note data, a content information table is constructed; the header of the content information table includes a note identifier, a note content score and content information; the content information includes at least one of the following: the number of note views, the length of the text, the number of pictures, the number of audios, the number of videos, the number of hyperlinks, and the number of files; the note content score is determined based on the content information and is used to indicate the probability of the user adding a note corresponding to the content information;

[0114] Determine the content information in the content information table as note features in the initial feature space corresponding to the content dimension;

[0115] Determine the evaluation index corresponding to the content dimension based on the note content score in the content information table.

[0116] Illustratively, the content information corresponding to the content dimension extracted from the local note data may include but is not limited to the number of views of each note, text length, number of pictures, number of audios, number of videos, number of hyperlinks, and number of files; it should be noted that based on different actual needs, the content information may include more or fewer types of information, and the present application does not limit this; the extracted information is filled into the table in units of notes to obtain a content information table. Taking the content information containing the above 7 types as an example, Table 4 shows the content information table provided by an exemplary embodiment of the present application. As shown in Table 4, the content information table contains 9 columns of data, and the corresponding table headers are note ID, note view count, text length, number of pictures, number of audios, number of videos, number of hyperlinks, number of files and note content score.

[0117] Table 4

[0118]

[0119] The note content score is calculated based on the content information and the first calculation formula. Schematically, the first calculation formula for the note content score can be expressed as:

[0120]

[0121] Among them, S1 is the note content score, t is the text length, c1 is the number of pictures, c2 is the number of audios, c3 is the number of videos, c4 is the number of hyperlinks, c5 is the number of files, and r is the number of views. Note content score is a standard to measure the richness of note content. The richer the content, the higher the score.

[0122] It should be noted that the calculation formula for the above-mentioned note content score is an example provided by this application. Based on different actual needs, the calculation formula or the parameters in the above-mentioned calculation formula can be adjusted, or the calculation formula can be customized, and this application does not impose any restrictions on this.

[0123] After obtaining the content information table, the local terminal device determines the evaluation index corresponding to the content dimension based on the note content score in the content information table, including:

[0124] Determine the expected value of the content score based on the note content score corresponding to each note in the content information table;

[0125] Based on the note content score, content score expectation value and normal distribution curve formula corresponding to each note, the content distribution score of each note is calculated;

[0126] The content distribution score of each note is determined as the value of the evaluation index of each note under the content dimension.

[0127] Indicatively, the local terminal device obtains the note content score corresponding to each note in the content information table, calculates the expected value μ of the content score, and sets get Among them, 100 means that the highest value of the note content score is 100;

[0128] The normal distribution curve formula constructed using the values ​​of μ and σ can be expressed as:

[0129]

[0130] Among them, x is the note content score, f(x) is the content distribution score (the highest score is 100), and the content distribution score of each note can be substituted into the above normal distribution curve formula to obtain the content distribution score corresponding to each note. Furthermore, the content distribution score corresponding to each note is added to the corresponding position of each note in the content information table; the content distribution score can represent the generation probability of the note content. The higher the content distribution score, the more likely the user is to add notes containing corresponding content information.

[0131] The content information in the content information table is used as the note features in the initial feature space under the content dimension, that is, {note viewing times, text length, number of pictures, number of audios, number of videos, number of hyperlinks, number of files} in the content information table is used as input parameters; further, the input parameters can be upgraded in dimension by polynomial regression to obtain an extended feature space. Schematically, taking the feature upgrade by the PolynomialFeatures tool of sklearn and degree = 2 as an example, the extended feature space under the classification dimension can be expressed as:

[0132] X=[[1.0,a,b,c,d,e,f,g,a*b,a*c,a*d,a*e,a*f,a*g,b*c,b*d,b*e,b*f,b*g,c*d,c*e,c*e,c*f,c*f,c*g,d*e,d*f,d*g,e*f,e*g,f*g]]

[0133] Among them, a is the number of note views, b is the length of the text, c is the number of pictures, d is the number of audios, e is the number of videos, f is the number of hyperlinks, and g is the number of files; the feature interactions between the various note features do not involve self-multiplication.

[0134] The content distribution score is used as an evaluation indicator corresponding to the content dimension, that is, the content distribution score is used as an output result.

[0135] The extended feature space X under the content dimension and the output result Y are used as training sets to train a linear regression model to obtain a content local model; this process can be performed using the StandardScaler and LinearRegression tools of sklearn.

[0136] The local terminal device exports the parameters of the content local model as content user features, and the number of parameters of the content local model corresponds to the number of features in the extended feature space under the content dimension; schematically, the number of parameters of the content local model corresponding to the content information table shown in Table 4 is 29, including: [0.80174849, 2.62729573, 1.37674657, 2.46598466, 0.70605734, -0.61203878, 0.91143223, 2.5518196, 1.80583183, 1.4729953, -1.94 918721,1.73895819,-0.3456492,0.82625111,-1.80474882,2.9564482,-1.60028575,-0.14336678,0.03226522,-2.38731489,-1.19900984,2.23537995,0.97661668,1.53676226,-1.22843199,-1.20480698,0.80337901,-1.88246088,-0.82708758].

[0137] 3. Editing Dimensions:

[0138] The editing dimension refers to the summary of the final results of the user's changes to the rich text format of the note. The editing local model generated under the editing dimension can measure the user's usage depth tendency (frequent or not) and complexity tendency (how much each time they use it) in the editing dimension.

[0139] The process of generating and editing a local model on a local terminal device can be implemented as follows:

[0140] Based on the editing information corresponding to the editing dimension extracted from the local note data, an editing information table is constructed, wherein the header of the editing information table includes a note identifier, a note editing score, and editing information; the editing information includes at least one of the following: the number of edits, font size, font color, emphasis style, typesetting style, list style, and table style; the note editing score is determined based on the editing information and is used to indicate the probability of the user adding a note corresponding to the editing information;

[0141] Determine the editing information in the editing information table as note features in the initial feature space corresponding to the editing dimension;

[0142] The evaluation index corresponding to the editing dimension is determined based on the note editing score in the editing information table.

[0143] Illustratively, the content information corresponding to the content dimension extracted from the local note data may include but is not limited to the number of edits of each note, font size, font color, emphasis style, typesetting style, list style, and table style; it should be noted that based on different actual needs, the content information may include more or fewer types of information, and the present application does not limit this; in the embodiment of the present application, the parameter corresponding to the font size is the total length of the text with a non-default font size; the parameter corresponding to the font color is the total length of the text with a non-default font color; the parameter corresponding to the emphasis style is the total length of the text with bold / strikethrough / underline set; the parameter corresponding to the typesetting style is an unexpected setting for the default style The total length of the left / center / right aligned text; the parameter corresponding to the list style is the total number of rows in the ordered / unordered list; the parameter corresponding to the table style is the total number of grids in the table, where, if there are merged cells, they are calculated before the cells are merged; the above information of each note extracted from the user's local note data is filled into the table in units of notes to obtain an editing information table. Taking the editing information including the above 7 types as an example, Table 5 shows the editing information table provided by an exemplary embodiment of the present application. As shown in Table 5, the editing information table contains 9 columns of data; the corresponding table headers are note ID, number of edits, font size, font color, emphasis style, typesetting style, list style, table style and note editing score.

[0144] Table 5

[0145]

[0146] The note editing score is calculated based on the editing information and the second calculation formula. Schematically, the second calculation formula for the note editing score can be expressed as:

[0147] S2=(x1+x2+x3+x4+5x5+2x6) / log2t

[0148] Among them, S2 is the note editing score, x1 is the font size, x2 is the font color, x3 is the emphasis style, x4 is the typesetting style, x5 is the list style, x6 is the table style, and t is the number of edits. The note editing score is a standard to measure the complexity of note editing. The more complex the editing, the higher the score.

[0149] It should be noted that the calculation formula for the above-mentioned note editing score is an example provided by this application. Based on different actual needs, the calculation formula or the parameters in the above-mentioned calculation formula can be adjusted, or the calculation formula can be customized, and this application does not impose any restrictions on this.

[0150] After obtaining the editing information table, the local terminal device determines the evaluation criteria for the editing dimension based on the note editing score in the editing information table, including:

[0151] Determine an expected editing score based on the note editing score corresponding to each note in the editing information table;

[0152] Based on the note editing score, editing score expectation value and normal distribution curve formula corresponding to each note, the editing distribution score of each note is calculated;

[0153] The editing distribution score of each note is determined as the value of the evaluation index of each note under the editing dimension.

[0154] Indicatively, the local terminal device obtains the note editing score corresponding to each note in the editing information table, calculates the expected value μ of the editing score, and sets get Among them, 100 means that the highest value of the note editing score is 100;

[0155] The normal distribution curve formula constructed using the values ​​of μ and σ can be expressed as:

[0156]

[0157] Among them, x is the note editing score, f(x) is the editing distribution score (the highest score is 100), and the editing distribution score corresponding to each note can be obtained by substituting the note editing score of each note into the above normal distribution curve formula. Furthermore, the editing distribution score corresponding to each note is added to the corresponding position of each note in the editing information table; the editing distribution score can represent the generation probability of notes with the corresponding editing method. The higher the editing distribution score, the more likely the user is to add notes with the corresponding editing method.

[0158] The editing information in the editing information table is used as the note features in the initial feature space under the editing dimension, that is, {edit times, font size, font color, emphasis style, typesetting style, list style, table style} in the editing information table is used as input parameters; further, the input parameters can be upgraded by polynomial regression to obtain an extended feature space. Schematically, taking the feature upgrade by the PolynomialFeatures tool of sklearn and degree = 2 as an example, the extended feature space under the classification dimension can be expressed as:

[0159] X=[[1.0,a,b,c,d,e,f,g,a*b,a*c,a*d,a*e,a*f,a*g,b*c,b*d,b*e,b*f,b*g,c*d,c*e,c*e,c*f,c*f,c*g,d*e,d*f,d*g,e*f,e*g,f*g]]

[0160] Among them, a is the number of edits, b is the font size, c is the font color, d is the emphasis style, e is the typesetting style, f is the list style, and g is the table style. The feature interactions between the various note features do not involve self-multiplication.

[0161] The editing distribution score is used as an evaluation indicator corresponding to the editing dimension, that is, the editing distribution score is used as an output result.

[0162] The extended feature space X under the editing dimension and the output result Y are used as training sets to train a linear regression model to obtain the edited local model; this process can be performed using the StandardScaler and LinearRegression tools of sklearn.

[0163] The local terminal device exports the parameters of the edited local model as the edited user features, and the number of the parameters of the edited local model corresponds to the number of features in the extended feature space under the edit dimension; schematically, the number of parameters of the edited local model corresponding to the edit information table shown in Table 5 is 29, including: [-1.63510124, 0.93557997, 1.43687535, 0.51427204, -1.31817204, 2.72292264, 1.45857776, 2.39239927, -1.569865, 2.47015921, 1 .07320353,1.43155687,-0.27669296,1.19577973,-1.16458401,0.60416494,0.27368671,1.35190302,1.95616182,0.74278312,-1.03447527,2.94330578,-0.0060608,-2.88144912,1.11492212,2.92108627,1.57158602,2.67832838,0.71899055].

[0164] Since the user's local note data will continue to change during use, if the user feature generation process (i.e., the local model training process) is re-executed after the local note data changes, a large amount of local resources will be consumed; therefore, in order to optimize the scenario where local note data changes frequently, when the data changes, the corresponding information table can be updated first, that is, the note information table, content information table and editing information table can be updated, and the updated mark can be made.

[0165] When the model training node is reached, if the target dimension is a classification dimension, the process of note feature extraction and linear regression model training is performed based on the updated classification information table to obtain an updated classification local model. The updated classification information table is determined based on the updated note information table;

[0166] When the model training node is reached, if the target dimension is the content dimension or the editing dimension, the note feature extraction and linear regression model training process is performed based on the updated information in the updated content information table or the editing information table to obtain an updated content local model or editing local model.

[0167] That is to say, when the local note data is updated, the classification local model generation process corresponding to the classification dimension is fully executed to obtain the updated classification local model, so as to obtain the updated classification user characteristics; when the content local model corresponding to the content dimension is updated, the existing content local model is reinforced trained based on the updated information in the content information table to obtain the updated content local model, so as to obtain the updated content user characteristics; when the editing local model corresponding to the editing dimension is updated, the existing editing local model is reinforced trained based on the updated information in the editing information table to obtain the updated editing local model, so as to obtain the updated editing user characteristics, thereby shortening the time required for model training, improving the model training efficiency, and saving local computing resources.

[0168] The model training node can be a user feature output node set based on actual needs; it should be noted that the training process of the local model of the above-mentioned dimensions can be executed in whole or in part based on actual needs; illustratively, if the local model training is performed on the above-mentioned three dimensions based on the local note data, the user features in the three dimensions are obtained, and the user features can be output; taking the user features illustratively shown in the above-mentioned embodiment as an example, Table 6 shows the output format of the user features provided by an exemplary embodiment of the present application.

[0169] Table 6

[0170]

[0171]

[0172] Step 250, perform user feature analysis based on the local model of the target dimension to obtain analysis results; the analysis results are used to indicate the user's note usage tendency in the target dimension.

[0173] In one possible implementation, after obtaining the local model of the target dimension, the user's note usage tendency can be analyzed based on the user characteristics reflected by the local model; the note usage tendency can refer to the user's usage preference for note operations. For example, through analysis, the user's commonly used fonts, or commonly used editing combinations, etc. can be determined. By analyzing the user's note usage tendency, it is convenient to provide users with personalized product services, thereby improving the user's product usage experience.

[0174] The analysis process can be implemented as follows:

[0175] Perform user feature analysis based on model parameters in the local model of the target dimension to obtain analysis results;

[0176] In one possible implementation, since the local model is a linear model, each coefficient of the linear model corresponds to each feature. Therefore, the coefficient performance based on the interactive feature can analyze the relationship between any two or more input parameters; in another possible implementation, the user preference analysis is performed by constructing input parameters, that is, by fixing other parameters and setting a unique variable to analyze the user's inclination towards the variable; schematically, other parameters are fixed in the content local model, and the text length is set as the only variable. By inputting the increasing text length into the content local model, the user's inclination towards text length is analyzed based on the change in the obtained content distribution score. In another possible implementation, by comparing and analyzing the local models of different users, common characteristics of user groups can be obtained, such as whether the table function of note editing is frequently used, whether users will use the table function in depth, and so on.

[0177] Alternatively, the analysis process can be implemented as:

[0178] A corresponding two-dimensional image is generated based on the local model of the target dimension, so as to perform user feature analysis based on the two-dimensional image and obtain analysis results.

[0179] In this case, the local terminal device can construct a mathematical expression based on the model parameters of the local model. Taking the classification local model as an example, the corresponding mathematical expression can be expressed as:

[0180]

[0181] Among them, f(x) is the total number of views, which can be regarded as the user's usage preference, k is the model parameter, x1 is the number of notes, and x2 is the number of tags. Based on the above mathematical expression, the corresponding two-dimensional image is drawn, and the relationship between the user's preference and each feature can be analyzed based on the two-dimensional image; for example, the x1 input is fixed to 100, and x2 and f(x) are used as the coordinate axes to draw a two-dimensional image, from which the relationship between the number of tags and the user's viewing tendency can be analyzed. For example, when the number of tags is 10, the conclusion can be obtained from the image that the user views the most notes.

[0182] Alternatively, analysis can also be performed based on the coefficients in the above mathematical expression. For example, if the value of coefficient k2 is less than 0, it means that the number of notes is negatively correlated with the number of views, and if it is greater than 0, it means that the correlation is positive. The larger the parameter value, the greater the impact.

[0183] When performing group commonality analysis, you can compare model parameters of different users horizontally. For example, to determine whether the table function of the editing dimension is frequently used, you can perform statistical processing on the parameters k8, k14, k19, k23, k26, k28, and k29 related to the editing dimension model and table style features, and calculate the mathematical expectation. If the parameter is greater than 0 and the value is larger, it means that the user group uses the table function more frequently.

[0184] For the content local model, the corresponding mathematical expression can be expressed as:

[0185] f(x)=k1+k2x1+k3x2+k4x3+k5x4+k6x5+k7x6+k8x7

[0186] +k9x1x2+k 10 x1x3+k 11 x1x4+k 12 x1x5+k 13 x1x6+k 14 x1x7

[0187] +k 15 x2x3+k 16 x2x4+k 17 x2x5+k 18 x2x6+k 19 x2x7

[0188] +k 20 x3x4+k 21 x3x5+k 22 x3x6+k 23 x3x7

[0189] +k 24 x4x5+k 25 x4x6+k 26 x4x7

[0190] +k 27 x5x6+k 28 x5x7

[0191] +k 29 x6x7

[0192] Where f(x) is the content distribution score, which indicates the user's preference, k is the model parameter, x1 is the number of views, x2 is the length of the text, x3 is the number of pictures, x4 is the number of audios, x5 is the number of videos, x6 is the number of hyperlinks, and x7 is the number of files. The relationship between the user's preference and each feature can be analyzed in a similar way to the classification local model, which will not be repeated here.

[0193] For editing the local model, the corresponding mathematical expression can be expressed as:

[0194] f(x)=k1+k2x1+k3x2+k4x3+k5x4+k6x5+k7x6+k8x7

[0195] +k9x1x2+k 10 x1x3+k 11 x1x4+k 12 x1x5+k 13 x1x6+k 14 x1x7

[0196] +k 15 x2x3+k 16 x2x4+k 17 x2x5+k 18 x2x6+k 19 x2x7

[0197] +k 20 x3x4+k 21 x3x5+k 22 x3x6+k 23 x3x7

[0198] +k 24 x4x5+k 25 x4x6+k 26 x4x7

[0199] +k 27 x5x6+k 28 x5x7

[0200] +k 29 x6x7

[0201] Among them, f(x) is the edit distribution score, which indicates the user's preference, k is the model parameter, x1 is the number of edits, x2 is the font size, x3 is the font color, x4 is the emphasis style, x5 is the typesetting style, x6 is the list style, and x7 is the table style. The relationship between the user's preference and each feature can be analyzed in a similar way to the classification local model, which will not be repeated here.

[0202] In summary, the note-based user feature analysis method provided in the embodiment of the present application is that the local terminal device obtains the user's local note data, extracts the note information of the target dimension from the local note data, and constructs the corresponding initial feature space based on the extracted note information and obtains the corresponding evaluation index. After that, a linear regression model is trained based on the initial feature space and the evaluation index of the local note data to obtain a local model under the target dimension, and user feature analysis is performed based on the local model to obtain an analysis result indicating the user's note usage tendency under the target dimension. Through the above method, user feature analysis under the target dimension based on local note data can be implemented in the local terminal device, which has good universality in the scenario of frequently deleted and modified note data, can reduce interference from erroneous operations and invalid information, and thus improve the accuracy of user feature analysis; in addition, by performing user feature analysis on the local terminal device, the privacy protection effect of user data can be improved, and the security of user data can be improved.

[0203] Figure 3 FIG. 1 shows a block diagram of a user feature analysis device based on notes provided by an exemplary embodiment of the present application. The device can perform the following steps: Figure 1 or Figure 2 All or part of the steps of the embodiment shown, such as Figure 3 As shown, the device comprises:

[0204] The acquisition module 310 is used to acquire the user's local note data;

[0205] A construction module 320, configured to construct an initial feature space and an evaluation index corresponding to the target dimension based on the note information corresponding to the target dimension extracted from the local note data; the initial feature space includes each note feature corresponding to the target dimension;

[0206] A training module 330, configured to perform linear regression model training based on the initial feature space and the evaluation index to obtain a local model corresponding to the target dimension;

[0207] The analysis module 340 is used to perform user feature analysis based on the local model of the target dimension to obtain analysis results; the analysis results are used to indicate the user's note usage tendency in the target dimension.

[0208] In a possible implementation, the training module 330 includes:

[0209] A dimension-increasing submodule, used for performing feature dimension-increasing processing on each note feature in the initial feature space to obtain an extended feature space;

[0210] The training submodule is used to perform linear regression model training based on each note feature in the extended feature space and the evaluation index to obtain a local model corresponding to the target dimension.

[0211] In a possible implementation, the construction module 320 includes:

[0212] An information table construction submodule, for constructing an information table corresponding to the target dimension based on the note information corresponding to the target dimension extracted from the local note data; the information in the information table is updated in real time as the local note data changes;

[0213] The information determination submodule is used to determine the initial feature space and evaluation index corresponding to the target dimension based on the information table corresponding to the target dimension.

[0214] In a possible implementation, the target dimension includes a classification dimension, a content dimension, and an editing dimension;

[0215] The local models correspondingly include classification local models, content local models and editing local models.

[0216] In one possible implementation,

[0217] In a possible implementation, the information table construction submodule is used to:

[0218] In the case where the target dimension is a classification dimension, a note information table is constructed based on classification information corresponding to the classification dimension extracted from the local note data, wherein a header of the note information table includes a note identifier, a note tag array, and a note viewing count;

[0219] Based on the note information table, classification information is accumulated and counted to obtain a classification information table; the header of the classification information table includes the number of notes, the number of tags, and the total number of note views; wherein the number of tags is used to indicate the number of non-repeated tags between multiple notes;

[0220] The information determination submodule is used to:

[0221] Determine the number of notes and the number of tags in the note information table as note features in the initial feature space corresponding to the classification dimension;

[0222] The total note viewing times in the note information table is determined as an evaluation index corresponding to the classification dimension.

[0223] In a possible implementation, the information table construction submodule is used to:

[0224] In the case where the target dimension is a content dimension, a content information table is constructed based on the content information corresponding to the content dimension extracted from the local note data; the header of the content information table includes a note identifier, a note content score and the content information; the content information includes at least one of the following: the number of note views, the length of the text, the number of pictures, the number of audios, the number of videos, the number of hyperlinks, and the number of files; the note content score is determined based on the content information and is used to indicate the probability of a user adding a note corresponding to the content information;

[0225] The information determination submodule is used to:

[0226] Determining the content information in the content information table as note features in the initial feature space corresponding to the content dimension;

[0227] An evaluation index corresponding to the content dimension is determined based on the note content score in the content information table.

[0228] In a possible implementation, the information table construction submodule is used to:

[0229] In the case where the target dimension is an editing dimension, an editing information table is constructed based on the editing information corresponding to the editing dimension extracted from the local note data, the header of the editing information table including a note identifier, a note editing score and the editing information; the editing information includes at least one of the following: number of edits, font size, font color, emphasis style, typesetting style, list style and table style; the note editing score is determined based on the editing information and is used to indicate the probability of a user adding a note with corresponding editing information;

[0230] The information determination submodule is used to determine the editing information in the editing information table as a note feature in the initial feature space corresponding to the editing dimension;

[0231] An evaluation index corresponding to the editing dimension is determined based on the note editing score in the editing information table.

[0232] In a possible implementation manner, the device further includes:

[0233] The model updating module is used to perform the process of note feature extraction and linear regression model training based on the updated classification information table when the model training node is reached and the target dimension is the classification dimension, so as to obtain the updated classification local model.

[0234] In a possible implementation, the model updating module is used to:

[0235] When the model training node is reached and the target dimension is the content dimension or the editing dimension, the process of note feature extraction and linear regression model training is performed based on the updated information in the updated content information table or the editing information table to obtain the updated content local model or the editing local model.

[0236] In a possible implementation, the analysis module 340 is used to:

[0237] Performing user feature analysis based on model parameters in the local model of the target dimension to obtain the analysis result;

[0238] or,

[0239] A corresponding two-dimensional image is generated based on the local model of the target dimension, so as to perform user feature analysis based on the two-dimensional image to obtain the analysis result.

[0240] In summary, the note-based user feature analysis device provided in the embodiment of the present application is applied in a local terminal device, so that the local terminal device obtains the user's local note data, extracts the note information of the target dimension from the local note data, and constructs the corresponding initial feature space based on the extracted note information and obtains the corresponding evaluation index. After that, a linear regression model is trained based on the initial feature space and the evaluation index of the local note data to obtain a local model under the target dimension, and user feature analysis is performed based on the local model to obtain an analysis result indicating the user's note usage tendency under the target dimension. Through the above-mentioned device, user feature analysis under the target dimension based on local note data can be implemented in the local terminal device, which has good universality in the scenario of frequently deleted and modified note data, can reduce interference from erroneous operations and invalid information, and thus improve the accuracy of user feature analysis; in addition, by performing user feature analysis on the local terminal device, the privacy protection effect of user data can be improved, and the security of user data can be improved.

[0241] Figure 4The following is a block diagram of a computer device 400 according to an exemplary embodiment of the present application. The computer device 400 may be implemented as the above-mentioned local terminal device, such as a smart phone, a tablet computer, a notebook computer, a desktop computer, etc. The computer device 400 may also be referred to as a user device, a portable terminal, a laptop terminal, a desktop terminal, or other names.

[0242] Typically, the computer device 400 includes a processor 401 and a memory 402 .

[0243] In some embodiments, the computer device 400 may further optionally include: a peripheral device interface 403 and at least one peripheral device. The processor 401, the memory 402 and the peripheral device interface 403 may be connected via a bus or a signal line. Each peripheral device may be connected to the peripheral device interface 403 via a bus, a signal line or a circuit board. Specifically, the peripheral device includes: at least one of a radio frequency circuit 404, a display screen 405, a camera assembly 406, an audio circuit 407 and a power supply 408.

[0244] In some embodiments, the computer device 400 further includes one or more sensors 409 , including but not limited to: an acceleration sensor 410 , a gyroscope sensor 411 , a pressure sensor 412 , an optical sensor 413 , and a proximity sensor 414 .

[0245] Those skilled in the art will understand that Figure 4 The structure shown in the figure does not constitute a limitation on the computer device 400, and the computer device 400 may include more or less components than those shown in the figure, or combine some components, or adopt a different arrangement of components.

[0246] In an exemplary embodiment, a computer-readable storage medium is also provided, in which at least one computer program is stored, and the computer program is loaded and executed by a processor to implement all or part of the steps in the above-mentioned user feature analysis method based on notes. For example, the computer-readable storage medium can be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), a magnetic tape, a floppy disk, an optical data storage device, etc.

[0247] In an exemplary embodiment, a computer program product is also provided, the computer program product comprising at least one computer program, the computer program being loaded and executed by a processor Figure 1 or Figure 2All or part of the steps of the note-based user feature analysis method shown in any embodiment.

[0248] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the claims.

[0249] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A user feature analysis method based on notes, characterized in that: The method is performed by a local terminal device, and the method includes: Get the user's local note data; Based on the note information corresponding to the target dimension extracted from the local note data, construct an initial feature space and an evaluation index corresponding to the target dimension; the initial feature space contains each note feature corresponding to the target dimension; Performing linear regression model training based on the initial feature space and the evaluation index to obtain a local model corresponding to the target dimension; Perform user feature analysis based on the local model of the target dimension to obtain analysis results; the analysis results are used to indicate the user's note usage tendency in the target dimension.

2. The method according to claim 1, characterized in that The linear regression model training is performed based on the initial feature space and the evaluation index to obtain a local model corresponding to the target dimension, including: Performing feature dimension-upgrading processing on each note feature in the initial feature space to obtain an extended feature space; A linear regression model is trained based on each note feature in the extended feature space and the evaluation index to obtain a local model corresponding to the target dimension.

3. The method according to claim 1, characterized in that The constructing an initial feature space and an evaluation index corresponding to the target dimension based on the note information corresponding to the target dimension extracted from the local note data includes: Based on the note information corresponding to the target dimension extracted from the local note data, construct an information table corresponding to the target dimension; the information in the information table is updated in real time as the local note data changes; Based on the information table corresponding to the target dimension, an initial feature space and an evaluation index corresponding to the target dimension are determined.

4. The method according to any one of claims 1 to 3, characterized in that: The target dimensions include classification dimension, content dimension and editing dimension; The local models correspondingly include classification local models, content local models and editing local models.

5. The method according to claim 4, characterized in that In the case where the target dimension is a classification dimension, constructing an information table corresponding to the target dimension based on the information corresponding to the target dimension extracted from the local note data includes: Based on the classification information corresponding to the classification dimension extracted from the local note data, a note information table is constructed, wherein a header of the note information table includes a note identifier, a note tag array, and a note viewing count; Based on the note information table, classification information is accumulated and counted to obtain a classification information table; the header of the classification information table includes the number of notes, the number of tags, and the total number of note views; wherein the number of tags is used to indicate the number of non-repeated tags between multiple notes; The determining, based on the information table corresponding to the target dimension, an initial feature space and an evaluation index corresponding to the target dimension comprises: Determine the number of notes and the number of tags in the note information table as note features in the initial feature space corresponding to the classification dimension; The total note viewing times in the note information table is determined as an evaluation index corresponding to the classification dimension.

6. The method according to claim 4, characterized in that In the case where the target dimension is a content dimension, constructing an information table corresponding to the target dimension based on the information corresponding to the target dimension extracted from the local note data includes: Based on the content information corresponding to the content dimension extracted from the local note data, a content information table is constructed; the header of the content information table includes a note identifier, a note content score and the content information; the content information includes at least one of the following: the number of note views, the length of the text, the number of pictures, the number of audios, the number of videos, the number of hyperlinks, and the number of files; the note content score is determined based on the content information and is used to indicate the probability of a user adding a note corresponding to the content information; The determining, based on the information table corresponding to the target dimension, an initial feature space and an evaluation index corresponding to the target dimension comprises: Determining the content information in the content information table as note features in the initial feature space corresponding to the content dimension; An evaluation index corresponding to the content dimension is determined based on the note content score in the content information table.

7. The method according to claim 4, characterized in that In the case where the target dimension is an editing dimension, constructing an information table corresponding to the target dimension based on the information corresponding to the target dimension extracted from the local note data includes: Based on the editing information corresponding to the editing dimension extracted from the local note data, an editing information table is constructed, wherein the header of the editing information table includes a note identifier, a note editing score and the editing information; the editing information includes at least one of the following: the number of edits, font size, font color, emphasis style, typesetting style, list style and table style; the note editing score is determined based on the editing information and is used to indicate the probability of a user adding a note corresponding to the editing information; The determining, based on the information table corresponding to the target dimension, an initial feature space and an evaluation index corresponding to the target dimension comprises: Determining the editing information in the editing information table as a note feature in the initial feature space corresponding to the editing dimension; An evaluation index corresponding to the editing dimension is determined based on the note editing score in the editing information table.

8. The method according to claim 4, characterized in that In the case where the target dimension is a classification dimension, the method further includes: When the model training node is reached, the process of note feature extraction and linear regression model training is performed based on the updated classification information table to obtain the updated classification local model.

9. The method according to claim 4, characterized in that In the case where the target dimension is a content dimension or an editing dimension, the method further includes: When the model training node is reached, the process of note feature extraction and linear regression model training is performed based on the updated information in the updated content information table or the edit information table to obtain the updated content local model or the edit local model.

10. The method according to claim 1, characterized in that The user feature analysis is performed based on the local model of the target dimension to obtain analysis results, including: Performing user feature analysis based on model parameters in the local model of the target dimension to obtain the analysis result; or, A corresponding two-dimensional image is generated based on the local model of the target dimension, so as to perform user feature analysis based on the two-dimensional image to obtain the analysis result.

Images