A method for accurate recommendation of printing services based on multi-dimensional user behavior data

Abstract

This invention belongs to the field of printing service technology and discloses a method for accurate printing service recommendation based on multi-dimensional user behavior data. The method includes: constructing user profile tags and parsing file attribute data; obtaining the user's historical printing order information and quantifying preference features; generating final scene tags based on multi-dimensional scene data; filtering feasible basic printing parameters according to the file attribute data, final scene tags, and a preset business rule base; expanding and generating N candidate solutions based on feasible basic printing parameters and the preference features; determining the utility value of each candidate solution through utility function calculation; determining the recommended solution based on the utility value; generating presentation content based on the recommended solution and the decomposition result of the utility function and pushing it to the user; and generating a revised recommended solution based on the feedback information. This invention achieves highly personalized and accurate recommendations, improving user experience.

Description

Technical Field

[0001] This invention relates to the field of printing service technology, and more specifically, to a method for accurate recommendation of printing services based on multi-dimensional user behavior data. Background Technology

[0002] With the development of information technology, printing services have expanded widely from offline physical stores to online platforms. When users submit print jobs through online platforms, they face a series of complex choices, including paper type, size, color, single / double-sided printing, print quality, and binding method. Currently, most online printing service platforms only provide a standardized list of options for users to choose from. This approach has the following drawbacks:

[0003] High level of expertise required: For ordinary users, it is difficult to make the best choice if they do not understand the applicable scenarios, page requirements and presentation effects of various binding methods (such as perfect binding, saddle stitching, spiral binding, butterfly binding, hardcover binding, etc.).

[0004] Inefficient selection: Users need to filter through numerous options one by one, which is cumbersome and results in a poor user experience.

[0005] Insufficient personalization: The system is unable to provide personalized suggestions based on the user's specific file content, historical preferences, or potential needs, which may result in the user choosing a less-than-optimal solution and affecting the final document presentation quality.

[0006] Existing technologies also include some simple recommendation mechanisms, such as recommending binding methods based on the number of pages in a document. However, such rules are too simplistic and static, failing to comprehensively consider dynamic information such as the user's identity, document content, and historical behavior. Consequently, the accuracy of the recommendations is low, and they cannot meet users' needs for professional and personalized printing services. Summary of the Invention

[0007] To overcome the aforementioned deficiencies of the prior art and achieve the above objectives, the present invention provides the following technical solution:

[0008] A method for accurate recommendation of printing services based on multi-dimensional user behavior data, the method comprising:

[0009] S1. Collect registration information through the registration form, build user profile tags, receive the file to be printed, and parse the file attribute data;

[0010] S2. Obtain the user's historical print order information, clean it, and quantify the preference characteristics through frequency statistics;

[0011] S3. Collect multi-dimensional scene data in real time, and generate final scene labels based on the multi-dimensional scene data;

[0012] S4. Based on the file attribute data, final scene tags, and preset business rule base, feasible basic printing parameters are obtained. Based on the feasible basic printing parameters and the preference features, N sets of candidate schemes are generated. The utility value of each candidate scheme is determined by calculating the utility function. The recommended scheme is determined based on the utility value.

[0013] S5. Generate presentation content based on the decomposition results of the recommended scheme and utility function, and push it to the user. The presentation content includes a description of the recommended scheme, the reasons for the recommendation based on the utility function, visual display content, scheme selection, and a slider for customizing parameters.

[0014] S6. Receive user feedback on the presented content, and generate a revised recommendation scheme and current print order information based on the feedback; wherein the feedback includes explicit feedback and implicit feedback.

[0015] Furthermore, the process of collecting registration information through a registration form, constructing user profile tags, receiving the file to be printed, and parsing the file attribute data includes:

[0016] S1.1 Build a registration form interface, set required and optional fields, and collect registration information. The required fields include basic identity information and privacy authorization; the basic identity information includes user ID, mobile phone number, and identity tag.

[0017] S1.2. Generate user profile tags using the registration information. The user profile tags include identity tags.

[0018] S1.3 Receive the file to be printed and parse it to obtain the file attribute data.

[0019] Furthermore, the process of obtaining users' historical print order information, and then quantifying preference features through frequency statistics after cleaning, includes:

[0020] S2.1 Retrieve the user's historical print order information based on the user ID;

[0021] S2.2 Clean the historical printed order information and remove invalid data;

[0022] S2.3 Quantify preference characteristics through frequency statistics. Preference characteristics include parameter preference, price sensitivity, and timeliness preference.

[0023] Furthermore, the real-time acquisition of multi-dimensional scene data, and the generation of final scene tags based on the multi-dimensional scene data, includes:

[0024] S3.1 Real-time collection of multi-dimensional scene data, including time data, device data, and geographic location data;

[0025] S3.2. Generate preliminary scene labels based on the data of each dimension of the multi-dimensional data of the scene. The preliminary scene labels include time dimension labels, device dimension labels and geographical location dimension labels, and configure a reliability score for each dimension of the preliminary scene label.

[0026] S3.3. Predefine the matching degree between each candidate scene and each preliminary scene label, and calculate the comprehensive score between each candidate scene and the preliminary scene label by combining the confidence score; wherein, the candidate scenes are obtained from the predefined candidate scene list;

[0027] S3.4. Combine the comprehensive score and the preset mapping rules between dimension labels and sub-scenes to determine the final scene label. The final scene label consists of the main scene and the sub-scene.

[0028] Furthermore, the process of obtaining feasible basic printing parameters based on the file attribute data, final scene tags, and a preset business rule base, and then expanding and generating N candidate schemes based on these feasible basic printing parameters and the preference features, and determining the utility value of each candidate scheme through utility function calculation, includes:

[0029] Load the business rule library, traverse the business rule library based on the file attribute data, and perform the first round of filtering; perform the second round of filtering based on the final scene label; merge the results of the two rounds of filtering, remove conflicts, generate feasible basic printing parameters, and form a list of feasible basic printing parameters.

[0030] Based on the preference characteristics, weights are assigned to each of the feasible basic printing parameters, and M initial schemes are selected from the feasible basic printing parameters by combining the improved Monte Carlo sampling method.

[0031] Calculate the difference between the M initial schemes, use a clustering algorithm to group the M initial schemes, select representative schemes from each group, and finally select N initial schemes as candidate schemes.

[0032] Calculate the preference matching score, cost score, timeliness score and scenario adaptability score for each candidate solution, and calculate the utility value of each candidate solution based on the utility function.

[0033] Further, the first round of filtering is performed by traversing the business rule base based on the file attribute data; the second round of filtering is performed based on the final scene tags; the results of the two rounds of filtering are merged, conflicts are removed, and feasible basic printing parameters are generated. The content of the feasible basic printing parameter list includes:

[0034] Traverse the business rule base, match the file attribute conditions in the applicable conditions, extract and summarize the parameter constraints corresponding to the matching rules to form the first round of filtering result set; wherein, the business rule base includes rule ID, applicable conditions, parameter constraints and priority;

[0035] Based on the final scenario label, the business rule library is traversed to match the scenario conditions in the applicable conditions; the parameter constraints corresponding to the matching rules are extracted and summarized to form the second round of filtering result set.

[0036] The parameter constraints of the two rounds of screening result sets are merged, the required parameters and recommended parameters are retained, and the prohibited parameters are removed to form an initial merged set;

[0037] Identify conflict types, including numerical conflicts, range conflicts, and logical conflicts, and resolve conflicts based on the priorities mentioned above;

[0038] After integrating and resolving the parameter constraints, the parameters are categorized and organized according to paper type, page size, color mode, single / double-sided printing, binding method, and delivery time to obtain a list of feasible basic printing parameters.

[0039] Furthermore, the specific steps for selecting candidate solutions include:

[0040] M initial schemes are encoded as feature vectors, and the Euclidean distance between the initial schemes is calculated as the dissimilarity degree to construct an M×M scheme dissimilarity matrix. The number of cluster groups K is determined according to the number of initial schemes. The M schemes are divided into K clusters using the K-means algorithm. For each cluster, the L schemes closest to the cluster center are selected as representative schemes. A representative scheme is selected from each cluster to form an initial candidate scheme set. The distribution of the initial candidate scheme set in key dimensions is checked, and representative schemes are added from the corresponding clusters based on the distribution. Key dimensions include delivery time and binding method. The process is iteratively supplemented until N candidate schemes are generated.

[0041] Furthermore, the selection of M initial schemes from the feasible basic printing parameters using the improved Monte Carlo sampling method includes: randomly selecting a value for each parameter category from the feasible basic printing parameters, but the selection probability is proportional to the assigned weight, and repeating the sampling until M initial schemes are generated.

[0042] Furthermore, the content generated based on the decomposition results of the recommendation scheme and utility function includes:

[0043] S5.1 Generate a description of the recommended scheme based on the printing parameters of the recommended scheme;

[0044] S5.2. Generate recommendation reasons based on the scores of each item in the utility function to obtain recommendation reasons based on the utility function;

[0045] S5.3, A display diagram of printing parameters based on the recommended scheme, wherein the visual display content includes the display diagram.

[0046] Furthermore, the step of receiving user feedback on the presented content and generating a revised recommendation scheme and current print order information based on the feedback includes:

[0047] S6.1 Obtain user feedback, including solution confirmation and parameter customization;

[0048] S6.2 Obtain implicit feedback from users, including the duration of time they remain on displayed content and their order abandonment behavior;

[0049] S6.3 When the user clicks on the solution confirmation in the display feedback, the current print order information is automatically generated based on the solution selected by the user;

[0050] S6.4 When no solution confirmation is found in the displayed feedback, the recommended solution and presentation content are revised based on the displayed feedback and implicit feedback, and pushed to the user again until the preset conditions for stopping the push are met.

[0051] The technical effects and advantages of the present invention, a method for accurate recommendation of printing services based on multi-dimensional user behavior data, are as follows:

[0052] This invention comprehensively utilizes user identity information (user profile), preference features (parameter preferences, price sensitivity, timeliness preferences), and multi-dimensional scenario data (time, location, device) to construct a comprehensive user understanding model. This enables recommendation schemes to deeply align with users' identity characteristics, long-term habits, and current needs. By collecting and analyzing time, geographical location, and device information in real time, final scenario tags are generated, allowing recommendations to dynamically adjust according to changes in the user's environment, achieving highly personalized and accurate recommendations, and enhancing the user experience.

[0053] Through a two-round screening process using a business rule base, combined with improved Monte Carlo sampling and clustering algorithms, candidate solutions that are both personalized and diverse are efficiently and intelligently generated from a massive number of possible parameter combinations, greatly reducing the cognitive burden on users who manually configure all printing parameters. A utility function is also introduced to comprehensively consider preference matching, cost, timeliness, and scenario adaptability, providing a quantitative basis for determining the primary and alternative recommendation solutions, avoiding subjective judgment, and ensuring that the recommended solutions achieve comprehensive optimization across multiple key dimensions. This achieves accurate recommendations while improving service efficiency and conversion rates. Attached Figure Description

[0054] Figure 1 The flowchart of a method for accurate recommendation of printing services based on multi-dimensional user behavior data provided by the present invention is shown below. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] Example 1

[0057] Please see Figure 1 As shown in the figure, this embodiment of the invention provides a method for accurate recommendation of printing services based on multi-dimensional user behavior data, the method comprising:

[0058] S1. Collect registration information through the registration form, build user profile tags, receive the file to be printed, and parse the file attribute data; the file attribute data includes file type, total number of pages, page size, content tags, and color mode;

[0059] S1.1 Build a registration form interface, set required and optional fields, and collect registration information. Required fields include basic identity information and privacy authorization. The basic identity information includes user ID, mobile phone number, and identity tag. The identity tag includes students, company employees, designers, and individual users, etc. Privacy authorization refers to the user signing a data collection authorization form, clarifying the scope of use of historical behavior and multi-dimensional scenario data, etc. Historical behavior includes historical print order information, which includes order number, user ID, file attribute data, printing parameters (specifically including color mode, single / double-sided, paper size (A3, A4, A5, B4, B5), binding method (supports perfect binding, saddle stitching, loose-leaf binding, stapler binding, etc.)), order status, payment amount, etc. The order status includes payment status, completion status, and cancellation status. Multi-dimensional scenario data includes time data, device data, and geographical location data, etc. Optional fields include common printing scenarios and common devices. Common printing scenarios include academic materials, business documents, design cases, and personal documents, etc. Common devices include PC / mobile / tablet.

[0060] S1.2. Generate user profile tags using the registration information. The user profile tags include identity tags. Extract the user's identity directly from the identity tag field of the registration form and establish identity tags.

[0061] S1.3 Receive the file to be printed and parse it to obtain file attribute data, including file type, total number of pages, page size, content tags, and color mode. The file type is obtained by reading the file extension, such as pdf, doc, docx, etc. The total number of pages is obtained by calling the tool API to get page number parameters. The page size is obtained by parsing the page settings parameters of the file to be printed, mapping them to A3 / A4 / A5 / B4 / B5 / Letter. The page settings parameters include page size parameters (width, height), etc. The steps for obtaining content tags include: constructing a keyword library, which is a multi-level keyword library containing academic keywords (graduation thesis, academic paper, research report, etc.). The document categories are: Business (bid documents, business plans, contracts, etc.), Design (design sketches, portfolios, renderings, etc.), and Personal (meeting minutes, resumes, diaries, etc.). Each content tag corresponds to at least 10 keywords. The document parses the filenames of the documents to be printed, removing special characters and numbers. It uses a text extraction API to obtain the main text of the documents, removes stop words and punctuation, and performs stemming. It counts the number of matches between the filenames and keywords of each content tag, and the number of matches between the main text and keywords of each content tag. When the total number of matching keywords between the filename and the main text for the same content tag is ≥2, it is marked as the corresponding content tag. If multiple content tags simultaneously meet the above conditions, the tag with the most matching keywords is selected. Color modes include color mode and black and white mode. When any RGB or CMYK color space is detected, or non-grayscale color elements are present, it is determined to be color mode. When all text and graphics are grayscale and there are no color elements, it is determined to be black and white mode.

[0062] S2. Obtain the user's historical print order information, clean it, and quantify the preference characteristics through frequency statistics; among which, the quantification dimensions of preference characteristics include parameter preference, price sensitivity, and timeliness preference;

[0063] S2.1 Retrieve the user's historical print order information based on the user ID;

[0064] S2.2 Clean the historical print order information and remove invalid data (remove historical print order information with payment status and cancellation status, payment amount ≤ 0 yuan, and duplicate order submissions (multiple submissions of the same file within 10 minutes).

[0065] S2.3 Quantify preference characteristics through frequency statistics. Preference characteristics include parameter preference, price sensitivity and timeliness preference. Specifically, the quantitative value of parameter preference is: for each printing parameter (such as binding method: perfect binding), count the number of times perfect binding is selected in the historical printing order information. Calculate by the number of selections / total number of valid orders × 100%. A score ≥ 60% is recorded as strong preference, 30-60% is recorded as medium preference, and < 30% is recorded as weak preference. Price sensitivity: The ratio of the average price of historical print orders to the average price of similar orders on the printing platform is used. A ratio ≤0.8 is considered high sensitivity, 0.8-1.2 is considered medium sensitivity, and ≥1.2 is considered low sensitivity. Similar orders refer to orders with identical content tags and consistent page counts (including 0-10 pages, 10-50 pages, 50-200 pages, etc.). Timeliness preference: Based on the percentage of historical print orders that selected expedited service, timeliness preference is determined as strong (≥40%), medium (10-40%), or weak (≤10%).

[0066] S3. Collect multi-dimensional scene data in real time, and generate final scene labels based on the multi-dimensional scene data;

[0067] S3.1 Real-time collection of multi-dimensional scene data, including time data, device data, and geographic location data, corresponding to the time dimension, device dimension, and geographic location dimension, respectively. The time data refers to the current time obtained from the server; the device data includes device type, which includes PC / mobile / tablet; the geographic location data includes scene type, which includes office buildings, campuses, residences, public places, etc., with the specific location accurate to the business district or campus area; the geographic location data is obtained based on the user's authorized IP address and mapped to the scene type through an IP database; the IP database is a commercial IP geographic location database, such as IP2Location, MaxMind GeoIP2, and other professional databases.

[0068] S3.2. Generate preliminary scene labels based on the data of each dimension of the multi-dimensional data of the scenario. The preliminary scene labels include time dimension labels, device dimension labels, and geographic location dimension labels, and assign a confidence score (0-1 points) to each dimension of the preliminary scene label. The time dimension labels include time period labels and date type labels. The current time obtained from the server is mapped to obtain time period labels and date type labels. The time period labels are divided into morning peak hours (7:00-9:00), office hours (9:00-17:30), off-duty hours (17:30-19:00), and nighttime hours (19:00-24:00). The date type labels are divided into weekdays, weekends, and public holidays. Dimensional labels also include special node labels (obtained through user marking), such as exam weeks and project deadlines; device dimension labels include fixed device labels and mobile device labels, which are generated based on device type, for example, fixed device labels are generated for PCs and mobile device labels are generated for mobile devices; geographic location labels are generated from the scene type in the geographic location data, such as residential, office buildings, etc.; the confidence level of geographic location dimension labels is the highest, followed by time dimension labels, and the lowest is device dimension labels; specifically: the confidence level of geographic location dimension labels is 0.9, the confidence level of time dimension labels is 0.7, and if special node labels exist, the confidence level of time dimension labels is increased to 0.9; the confidence level of device dimension labels is 0.5.

[0069] S3.3 Calculate the comprehensive score between each candidate scene and the initial scene label. The comprehensive score of each candidate scene = Σ(dimensional label matching degree × confidence level); where the dimension label matching degree refers to the correlation between the initial scene label of that dimension and the candidate scene; the candidate scenes are obtained from a predefined candidate scene list, which includes main scenes such as office scenes and personal scenes. The office scene is further subdivided into sub-scenes such as corporate office and home office, and the personal scene is further subdivided into sub-scenes such as family documents, personal creation, and life records; the matching degree between each candidate scene and each initial scene label is predefined. For example, the matching degree between office hours and office scenes is set to 1.0, and the matching degree between office hours and personal scenes is set to 0.2; the matching degree between residence and office scenes is set to 0.2, and the matching degree between residence and office scenes is set to 1.0; the matching degree between PC and office scenes is set to 0.5, and the matching degree between PC and personal scenes is set to 0.5.

[0070] S3.4. Obtain the candidate scenario with the highest comprehensive score as the main scenario; for example, if the confidence score of office hours in the time dimension label is 0.7, the confidence score of residential in the geographical location dimension label is 0.9, and the confidence score of PC in the device dimension label is 0.5, when the candidate scenario is an office scenario, its comprehensive score = (1.0×0.7)+(0.2×0.9)+(0.8×0.5)=1.52; when the candidate scenario is a personal scenario, its comprehensive score = (0.2×0.7)+(1.0×0.9)+(0.5×0.5)=1.29; therefore, the office scenario is selected as the main scenario; based on the preset mapping rules between dimension labels and sub-scenarios and the predefined sub-scenario library, sub-scenarios are determined, and the final scenario label is generated by combining the main scenario and sub-scenarios. The final scenario label = main scenario + sub-scenario, such as office scenario - working from home, personal scenario - emergency exam preparation, etc.

[0071] The steps for determining sub-scenes based on preset mapping rules between dimension labels and sub-scenes and a predefined sub-scene library include:

[0072] Prioritizing geographic location labels > time labels > device labels, we filter dimension labels with a confidence level ≥ 0.8 within the main scenario (e.g., if the main scenario is an office scenario, the geographic location label is residential (confidence level 0.9), and the time label is nighttime (confidence level 0.7)). We then match these labels against a predefined sub-scenario library to obtain sub-scenes. Each sub-scenario corresponds one-to-one with the main scenario. For example, the office scenario corresponds to sub-scenes such as corporate office, home office, and temporary office; the personal scenario corresponds to sub-scenes such as family documents, personal creation, and emergency printing. The mapping rules between dimension labels and sub-scenes are as follows: for example, office scenario + residential + PC + nighttime corresponds to home office; office scenario + office building + PC + office hours corresponds to corporate office; and personal scenario + campus + mobile device + exam week corresponds to personal emergency exam preparation.

[0073] S4. Based on the file attribute data, final scene tags and preset business rule base, feasible basic printing parameters are obtained. Based on the feasible basic printing parameters and the preference features, N sets of candidate schemes are generated. The utility value of each candidate scheme is determined by calculating the utility function. Based on the utility value, a recommended scheme is determined. The recommended scheme includes a main recommended scheme and alternative recommended schemes.

[0074] S4.1 Load the business rule base, traverse the business rule base based on the file attribute data, and perform a first round of filtering; perform a second round of filtering based on the final scene tags; merge the results of the two rounds of filtering, remove conflicts, and generate feasible basic printing parameters to form a feasible basic printing parameter list. The parameter categories in the feasible basic printing parameter list include paper type (such as coated paper, matte paper, etc.), page size (such as A4, A3, etc.), color mode (such as color, black and white), single-sided or double-sided (such as single-sided, double-sided), and binding method (such as perfect binding, saddle stitching, etc.). The business rule base is stored in JSON or graph database format and includes rule ID, applicable conditions, parameter constraints, and priority. The applicable conditions include file attribute conditions generated by the rule (such as total number of pages, content tags, page size, color mode) and scene conditions (such as office scene - enterprise office, personal scene - family document). The parameter constraints include required parameters, recommended parameters, and prohibited parameters. The priority is an integer from 1 to 100, with higher values ​​indicating higher priority, used for conflict resolution.

[0075] The specific steps for traversing the business rule base based on the file attribute data to perform a first round of filtering; performing a second round of filtering based on the final scene tags; merging the results of the two rounds of filtering, removing conflicts, generating feasible basic printing parameters, and constructing a list of feasible basic printing parameters include:

[0076] The first round of screening involves: traversing the business rule base and matching the file attribute conditions (such as total number of pages, content tags, page size, and color mode) in the applicable conditions; extracting the parameter constraints corresponding to the matching rules, for example, for the rule that content tag = academic paper + total number of pages ≥ 50, extracting the required parameters: binding method = perfect binding, recommended parameters: paper type = matte paper, and prohibited parameters: saddle stitching; summarizing the parameter constraints of all matching rules to form the first round of screening result set;

[0077] The second round of filtering: Based on the final scenario tags, the business rule base is traversed to match the scenario conditions in the applicable conditions (such as office scenario - enterprise office, personal scenario - family documents); the parameter constraints corresponding to the matching rules are extracted, for example, the recommended parameters for the "office scenario - enterprise office" rule are extracted: color mode = color, and the prohibited parameter is: single-sided printing. The parameter constraints of all matching rules are summarized to form the second round of filtering result set;

[0078] The parameter constraints of the two rounds of screening result sets are merged, the required parameters and recommended parameters are retained, and the prohibited parameters are removed to form an initial merged set;

[0079] Identify conflict types, including numerical conflicts, range conflicts, and logical conflicts. Numerical conflicts refer to the same parameter being constrained by different rules to mutually exclusive values, such as rule A requiring color mode = color, while rule B requires color mode = black and white. Range conflicts refer to contradictions in the constraints on the range of parameter values, such as rule A requiring perfect binding when the total number of pages is ≥50, while rule B requires saddle stitching when the number of pages is ≤100. Logical conflicts refer to contradictions in the business logic of printing parameters, such as requiring both single-sided printing and double-sided binding to exist simultaneously.

[0080] Priority-based resolution: Take the parameter corresponding to the rule with the higher priority value among the conflict rules (priority 1-100, the higher the value, the higher the priority); if the priorities are the same, file attribute rules > scenario rules are used to determine the conflict; the system automatically selects the parameter constraints corresponding to the file attribute rules (such as rule A requiring color printing based on file content tags) and ignores the conflict constraints in the scenario rules (such as rule B requiring black and white printing based on office scenarios).

[0081] The decomposed parameter constraints are integrated and sorted by paper type, page size, color mode, single or double-sided printing, binding method, and delivery time to obtain a list of feasible basic printing parameters;

[0082] S4.2. Assign weights to each of the feasible basic printing parameters according to the preference features, and select M initial schemes from the feasible basic printing parameters by combining the improved Monte Carlo sampling method, where M>N;

[0083] Specifically, assigning weights to each feasible basic printing parameter based on the preference features means: extracting the parameter preference of each printing parameter from the preference features, mapping the parameter preference to a numerical weight (such as 0.8, 0.5, and 0.2), thereby achieving weight assignment for each feasible basic printing parameter;

[0084] The process of selecting M initial schemes from the feasible basic printing parameters using the improved Monte Carlo sampling method specifically refers to: randomly selecting a value for each parameter category (such as binding method and paper type) from the feasible basic printing parameters, but the selection probability is proportional to the assigned weight; for example, perfect binding has a higher weight and is more likely to be selected; sampling is repeated until M initial schemes are generated, where M > N, to ensure diversity;

[0085] S4.3 Calculate the difference between the M initial schemes, use a clustering algorithm (such as K-means) to group the M initial schemes, select representative schemes from each group, and finally select N initial schemes (such as N=10) as candidate schemes; each candidate scheme is obtained by combining printing parameters, for example: Scheme 1: {Paper type=matte paper, page size=A4, color mode=color, single / double-sided=double-sided, binding method=perfect binding, delivery time=standard}; Scheme 2: {Paper type=plain paper, page size=A4, color mode=black and white, single / double-sided=single-sided, binding method=saddle stitching, delivery time=express};

[0086] The specific steps for selecting candidate schemes include: encoding the M initial schemes into feature vectors, calculating the Euclidean distance between the initial schemes as the dissimilarity, and constructing an M×M scheme dissimilarity matrix; determining the number of cluster groups K=min(5,M / 10) based on the number of initial schemes, rounding up if the result is a non-integer; dividing the M schemes into K clusters / groups using the K-means algorithm; for each cluster, selecting the L schemes closest to the cluster center as representative schemes, where L is a positive integer greater than 1, which can be 2 or 3; first selecting a representative scheme from each cluster (i.e., selecting the scheme closest to the cluster center from the L representative schemes of each cluster) to form the initial... A candidate solution set (of size K) is generated. The distribution of the initial candidate solution set across key dimensions is checked: if some key dimension categories are under-covered (i.e., the number of solutions in that category is lower than the ideal number), representative solutions are added from the corresponding clusters. Key dimensions include delivery time, binding method, etc. This process iterates until the generated N sets of candidate solutions meet preset requirements in terms of key dimension distribution. The preset requirement is to ensure that the final N sets of candidate solutions are diverse and representative, making the distribution of candidate solutions as uniform as possible across dimensions such as delivery time and binding method. The ideal number is determined based on the total number of candidate solution sets N. Ideally, each key dimension category should be distributed as uniformly as possible. For example, if delivery time has 3 categories, the ideal number for each category is approximately N / 3 (rounded down). K-means clustering is used; this is existing technology and will not be elaborated upon here.

[0087] The specific steps for encoding the M initial schemes into feature vectors include:

[0088] Using the parameter categories in the feasible baseline parameter list as feature dimensions, define the value range for each dimension, for example:

[0089] Dimension 1: Paper type (Values: Plain paper = 0, Matte paper = 1, Coated paper = 2);

[0090] Dimension 2: Page size (values: A4=0, A3=1, A5=2, B4=3, B5=4);

[0091] Dimension 3: Color Mode (Values: Black and White = 0, Color = 1);

[0092] Dimension 4: Single-sided or double-sided (Values: Single-sided = 0, Double-sided = 1);

[0093] Dimension 5: Binding method (Values: stapled = 0, saddle stitching = 1, perfect binding = 2, hardcover = 3, loose-leaf binding = 4);

[0094] Dimension 6: Delivery Time (Values: Normal = 0, Standard = 1, Rush = 2);

[0095] Dimension 7: Cost (standardized value: 0-1);

[0096] For discrete parameters (such as paper type and binding method): convert them to integer codes according to a predefined mapping relationship; for continuous parameters (such as cost): use Min-Max standardization to map parameter values ​​to the [0,1] interval (standardized value = (actual value - minimum value) / (maximum value - minimum value)); where, actual value: refers to the specific value of the parameter in the current candidate solution, for example, the cost of the solution is 15 yuan; minimum value: refers to the minimum value of the parameter among all candidate solutions. For example, the minimum cost among all solutions is 10 yuan; maximum value: refers to the maximum value of the parameter among all candidate solutions, for example, the maximum cost among all solutions is 30 yuan; if the maximum value is equal to the minimum value (i.e., the parameter values ​​are the same in all solutions), then the standardized value is uniformly set to 0.5 to avoid division by zero error;

[0097] For each initial scheme, extract the quantized values ​​corresponding to each parameter in the order of dimension 1 to dimension 7, and combine them into a one-dimensional vector. For example, the feature vector corresponding to the scheme {plain paper, A4, black and white, double-sided, perfect binding, standard, cost 0.3} is [0,0,0,1,2,1,0.3]. Repeat the above steps to complete the encoding of M initial schemes.

[0098] S4.4 Calculate the preference matching score between each candidate solution and the preference feature. Calculate the cost score, timeliness score, and scenario adaptability score of each candidate solution based on the printing parameters. Calculate the utility value of each candidate solution based on the utility function. The utility function is a weighted sum of the preference matching score, cost score, timeliness score, and scenario adaptability score. The weight coefficients of each score in the utility function can be allocated and adjusted based on the preference feature and satisfy the normalization condition.

[0099] The calculation process for the preference matching score includes:

[0100] The quantified value of the parameter preference is used as the matching weight of each printing parameter in the candidate scheme. For example, if the quantified value of the parameter preference for perfect binding is 0.8, then the binding method dimension weight is 0.8.

[0101] For each printing parameter of the candidate solution, if it matches the preference parameter (e.g., the user strongly prefers perfect binding, so the binding method of the solution is perfect binding), then the matching score of that parameter = weight × 1.0; if it does not match, then the matching score of that parameter = weight × 0.1. The matching scores of all printing parameters are summed and normalized to the interval [0, 100] to obtain the preference matching score, which is calculated as: Preference Matching Score = (Total matching scores of printing parameters / Maximum possible total score) × 100. The maximum possible total score = Σ(weight of each parameter) × 1.0. The preference parameter refers to the printing parameter preference value quantified from the user's historical behavior, specifically derived from the frequency statistics results in S2.3. For example, if the user's historical printing orders show perfect binding as the preferred option 70% of the time, then the parameter preference quantification value for perfect binding is 0.8 (strong preference), and perfect binding is the preference parameter. The maximum possible total score refers to the upper limit of the score under ideal conditions (where all printing parameters of the candidate solution are completely consistent with the user's preference parameters). The calculation method is as follows: for each printed parameter, take its weight (parameter preference quantification value), multiply it by 1.0 (for perfect matching), and then sum them up;

[0102] The calculation process of cost score:

[0103] The pricing database of the printing platform is accessed to extract the unit cost of each printing parameter, such as coated paper unit price = 0.2 yuan / page, color printing = 0.15 yuan / page, perfect binding = 5 yuan / book;

[0104] The total cost is calculated based on the combination of printing parameters for the candidate schemes. Total cost = Σ (unit cost of printing parameters × usage) + fixed costs, such as paper unit price × total number of pages + binding fee + delivery fee.

[0105] Based on the cost range of similar orders (same content tags + consistent page count), the following scoring rules are set: total cost ≤ average price of similar orders × 0.8, corresponding to a cost score of 90-100; total cost between 0.8 and 1.2 times the average price of similar orders, corresponding to a cost score of 70-89; total cost ≥ 1.2 times the average price of similar orders, corresponding to a cost score of 0-69. The cost score of this solution is obtained by mapping according to the above scoring rules.

[0106] The calculation process of the timeliness score:

[0107] Convert the delivery time of candidate solutions (e.g., standard, rush) into specific hours (e.g., standard = 48 hours, standard = 24 hours, rush = 6 hours), and record it as follows. ;

[0108] Extract timeliness preferences (strong, medium, weak) from user preference characteristics and map them to expected hours, for example: strong = 4 hours, medium = 12 hours, weak = 24 hours, and record them as follows. ;

[0109] Calculating timeliness score based on exponential decay function The formula is as follows:

[0110] ;

[0111] in, This represents the timeliness score, ranging from (0, 100]. It is the natural constant (approximately 2.71828); the formula ensures that the smaller the difference, the higher the score, and when the difference is 0, the score is 100; as the difference increases, the score decreases exponentially.

[0112] The calculation process of scene adaptability score:

[0113] The recommended parameters corresponding to the final scenario label are extracted from the business rule base, such as color printing for enterprise office scenarios. For each printing parameter of the candidate solution, it is checked whether it conforms to the recommended parameters corresponding to the final scenario label. If it does, the conformity score of this printing parameter is determined to be 1.0; if it does not, the conformity score of this parameter is determined to be 0.1. The scenario adaptability score is calculated by weighted summation, and the formula is: Scenario adaptability score = (Σ printing parameter conformity score / total number of printing parameters) × 100; the total number of printing parameters is usually 6.

[0114] S4.5 Select the candidate scheme with the highest utility value as the primary recommended scheme, and select the candidate scheme with the second highest utility value as the alternative recommended scheme. The alternative recommended schemes are usually 2-3 sets. The primary recommended scheme and the alternative recommended schemes constitute the recommended scheme, which includes the printing parameters of the primary recommended scheme and the alternative recommended schemes.

[0115] S5. Based on the decomposition results of the recommended scheme and utility function, generate presentation content and push it to the user. The presentation content includes a description of the recommended scheme, the reasons for the recommendation based on the utility function, visual display content, scheme selection, and a parameter customization slider. The scheme selection is a confirmation button for providing the user with a recommended scheme, and the recommended schemes are sorted based on utility values. The parameter customization slider is an interactive control embedded in the presentation content, allowing the user to adjust certain printing parameters (such as paper type, binding method, and delivery time) in real time within the range of feasible basic printing parameters. It also simultaneously triggers front-end components that recalculate utility values ​​and refresh the presentation content. Its core function is to meet the user's personalized fine-tuning needs. It enhances the transparency and persuasiveness of the recommendation, helps users make quick decisions, and improves order conversion rates.

[0116] S5.1 Generate a description of the recommended scheme based on the printing parameters of the recommended scheme;

[0117] S5.1.1 Extract the specific values ​​of 6 core printing parameters from the recommended scheme (e.g., paper type = matte paper, binding method = perfect binding, delivery time = expedited), sort them in a fixed order of paper type, page size, color mode, single / double-sided, binding method, and delivery time, and generate a list of parameter key-value pairs; associate them with the user profile tags in S1.2 (e.g., students, corporate employees) and the final scenario tags in S3.4 (e.g., office scenario - corporate office, personal scenario - emergency exam preparation).

[0118] S5.1.2. A pre-built multi-dimensional language template library is constructed. The language template library contains a pre-established tag combination index (primary key: user profile tag + final scene tag, value: template ID). Templates in the language template library are classified according to the combination of user profile tag + final scene tag, with each combination corresponding to 1-2 templates, as shown in the example below:

[0119] Template 1 (Enterprise Employees + Office Scenarios - Enterprise Office): This is the preferred business solution. It is recommended to use {paper type} to print {page size} documents, {color mode} for double-sided printing + {binding method}, and {delivery time} for delivery. It is suitable for business reports / contract signing scenarios.

[0120] Template 2 (Student + Personal Scenario - Emergency Exam Preparation): This is an emergency exam preparation plan. We recommend printing files with the following paper types, page sizes, and color modes. Print on one or two sides (single or double-sided) and use the following binding method. Expedited delivery is available to meet the needs of quickly printing exam materials during exam week.

[0121] User profile tags and final scenario tags are combined to form a target tag combination. Based on the target tag combination, the tag combination index is queried. If a completely matching primary key exists, the corresponding template ID is directly obtained and marked as the optimal template. If no completely matching primary key exists, a general template is used as the optimal template. The general template recommends {paper type}{color mode}{single / double-sided} printing {page size} file, {binding method} processing, and {delivery time} delivery. A recommendation scheme description is generated based on the matching optimal template.

[0122] S5.2. Generate recommendation reasons based on the scores of each item in the utility function to obtain recommendation reasons based on the utility function;

[0123] Extract the scores of each item of the utility function (preference matching score, cost score, timeliness score, and scenario adaptability score); sort them from high to low to identify advantages (score > 80) and disadvantages (score < 60); generate recommendation reasons based on the utility function based on the advantages and disadvantages;

[0124] Reasons for recommending advantages: For each advantage, the core advantages are determined by ranking them according to preference matching degree > scenario adaptability > timeliness score > cost score. The number of core advantages is limited to a maximum of 3 to avoid information overload. Personalized reasons for recommending the core advantages are supplemented with historical printing order information (each score has a unique recommendation reason template, which can usually be supplemented and improved with historical printing order information, for example: preference matching degree score template: preference matching degree score is 90 points, 80% of your historical printing orders are perfect binding, and the printing parameters of this solution are completely in line with your usage habits); only the reasons for the core advantages are displayed.

[0125] Reasons for recommending weaknesses: For weaknesses (score < 60), provide an explanation of the reason for generation and optimization suggestions to avoid user confusion; for example: cost score (C < 60): cost score {C} (slightly higher than the average price of similar products), because you selected high-cost parameters, such as coated paper + expedited delivery; if you need to reduce costs, you can try changing the high-cost parameters to recommended alternative parameters, such as plain paper + standard delivery; if there are no weaknesses, omit this section; if there are multiple weaknesses, only the one with the lowest score will be displayed to avoid too much negative information;

[0126] S5.3, A display diagram showing the matching of printing parameters based on the recommended scheme; the visual display content includes the display diagram.

[0127] The display image is a picture / video content that presents the appearance of the final printed product. Its core function is to allow users to intuitively predict the printing result (such as binding style, paper texture, and size ratio). The steps to generate the display image include: extracting visual parameters from the printing parameters of the recommended scheme, including binding method, paper type, and page size; and extracting sample images that match the key parameters from the sample library and marking them as display images.

[0128] A sample library is a database or storage system containing multiple sample images and their metadata. Each sample image record includes an image ID; metadata includes binding method, paper type, page size, color mode, etc.; the metadata is encoded in the same way as the parameters in the business rule base, for example, the binding method uses the same integer encoding (perfect binding = 2).

[0129] The process of extracting sample images matching the key parameters from the sample library includes: assigning visual weights to each visual parameter to reflect its importance. For example, binding method may have the highest visual weight (0.5) because it directly affects the appearance; paper type is next (0.3); and page size has a lower visual weight (0.2). The total visual weights are 1. For each visual parameter, the metadata of the sample image is checked to see if it is exactly the same as the parameter of the recommended scheme. When they are the same, the similarity score = 1.0; when they are different, the similarity score = 0. The similarity score is calculated based on the similarity score and the visual weights. The similarity score = Σ(parameter's visual weight × parameter's similarity score). For example, when the binding method similarity score = 1.0, the paper type similarity score = 0 (not a match), and the page size similarity score = 1.0, the similarity score = (0.5 × 1.0) + (0.3 × 0) + (0.2 × 1.0) = 0.7. All sample images are sorted by similarity score, and the sample image with the highest similarity score is selected as the display image. The visual parameters that are the same as and different from the recommended scheme are marked in the display image.

[0130] S6. Receive user feedback on the presented content, and generate a revised recommendation scheme and current print order information based on the feedback; wherein the feedback includes explicit feedback and implicit feedback;

[0131] S6.1. Capture user display feedback through front-end interactive components, including scheme confirmation and parameter customization; scheme confirmation refers to the user clicking the confirmation button after selecting it; parameter customization refers to the user modifying the printing parameters by sliding the parameter customization slider.

[0132] S6.2. Capture implicit user feedback through the event tracking system, including the duration of time users spend viewing content and order abandonment behavior.

[0133] S6.3 When the user clicks to confirm the solution in the display feedback, a structured current print order information is automatically generated based on the solution selected by the user. The current print order information includes order metadata, file attribute data, printing parameters of the solution selected by the user, order amount, etc.

[0134] S6.4 When no solution confirmation is found in the displayed feedback (i.e., there is an order abandonment), the recommended solution and presentation content are corrected based on the displayed and implicit feedback, and pushed to the user again until a preset stop push condition is met. The preset stop push condition is: when the current print order information has been generated or the number of times the presentation content of the same printable file is pushed exceeds a preset number threshold within a unit time, the push to the user is stopped. For example, the preset number threshold can be 3 times, and the unit time can be 24 hours. The specific content of correcting the recommended solution and presentation content based on the displayed and implicit feedback includes: determining whether there is parameter customization modification in the displayed feedback; if so, capturing the user's dwell time on the parameter customization slider page; if the dwell time exceeds a preset time threshold, replacing the corresponding category print parameters in the recommended solution with the modified print parameters to generate the latest recommended solution; calculating the preference matching score, cost score, timeliness score, and scenario adaptability score of the latest recommended solution to obtain the latest utility value, and re-sorting the latest recommended solutions based on the latest utility value to correct the recommended solutions; generating the corresponding presentation content based on the corrected recommended solutions.

[0135] This invention's precise, considerate, and efficient recommendation experience can significantly improve users' overall satisfaction with printing service platforms, thereby enhancing user stickiness and promoting repeat purchases. Precise recommendations can reduce order cancellations, reprints, or customer complaints caused by users selecting inappropriate printing parameters, thus lowering operating costs. Simultaneously, understanding users' price sensitivity and timeliness preferences helps the platform implement differentiated service and pricing strategies.

[0136] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0137] All formulas in this manual are dimensionless and calculated numerically. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0138] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1.A method for accurate recommendation of printing services based on multi-dimensional user behavior data, characterized in that, The method comprises: S1, collecting registration information through a registration form, constructing a user portrait label, receiving a to-be-printed file, and analyzing to obtain file attribute data; S2, obtaining historical printing order information of the user, quantifying the preference features through frequency statistics after cleaning; S3, real-time collection of scene multi-dimensional data, and generation of final scene labels based on the scene multi-dimensional data; S4, filtering to obtain feasible basic printing parameters according to the file attribute data, the final scene label and the preset business rule library, and expanding to generate N sets of candidate schemes based on the feasible basic printing parameters combined with the preference features, calculating and determining the utility value of each candidate scheme through the utility function, and determining the recommended scheme based on the utility value; Wherein, the content of expanding to generate N sets of candidate schemes based on the feasible basic printing parameters combined with the preference features includes: According to the preference features, assign weights to each of the feasible basic printing parameters, select M sets of initial schemes from the feasible basic printing parameters by combining the improved Monte Carlo sampling method; calculate the difference between the M sets of initial schemes, group the M sets of initial schemes using a clustering algorithm, select a representative scheme from each group, and finally select N sets of initial schemes as candidate schemes; S5, generating presentation content based on the recommended scheme and the decomposition result of the utility function, and pushing it to the user, wherein the presentation content includes recommended scheme description, recommended reason based on the utility function, visual display content, scheme selection and parameter customization modification slider; S6, receiving feedback information of the user on the presentation content, and generating a revised recommended scheme and current printing order information based on the feedback information; wherein the feedback information includes explicit feedback and implicit feedback. 2.The method of claim 1, wherein, The content of collecting registration information through a registration form, constructing a user portrait label, receiving a to-be-printed file, and analyzing to obtain file attribute data includes: S1.1, build a registration form interface, set required fields and optional fields, collect registration information, required fields include basic identity information and privacy authorization; the basic identity information includes user ID, mobile phone number, identity label; S1.2, generate a user portrait label based on the registration information, the user portrait label includes an identity label; S1.3, receive a to-be-printed file, and analyze to obtain file attribute data. 3.The method of claim 2, wherein, The content of obtaining historical printing order information of the user, quantifying the preference features through frequency statistics after cleaning includes: S2.1, based on the user ID, call the historical printing order information of the user; S2.2, clean the historical printing order information, and remove invalid data; S2.3, quantifying the preference features through frequency statistics, the preference features include parameter preference, price sensitivity and timeliness preference. 4.The method of claim 1, wherein, The content of real-time collection of scene multi-dimensional data, and generation of final scene labels based on the scene multi-dimensional data includes: S3.1, real-time collection of scene multi-dimensional data, the scene multi-dimensional data includes time data, device data and geographic location data; S3.2, generating a preliminary scene label based on the data of each dimension of the scene multidimensional data, the preliminary scene label including a time dimension label, a device dimension label, and a geographic location dimension label, and each dimension of the preliminary scene label being assigned a confidence score; S3.3, predefining a matching degree between each candidate scene and each preliminary scene label, combining the confidence score to calculate a comprehensive score between each candidate scene and the preliminary scene label; wherein the candidate scene is obtained from a pre-defined candidate scene list; S3.4, determining a final scene label in combination with the comprehensive score and a preset mapping rule of the dimension label and the sub-scene, the final scene label being composed of a main scene and a sub-scene. 5.The method of claim 1, wherein, The content of filtering the feasible basic printing parameters according to the file attribute data, the final scene label, and the preset business rule library includes: loading the business rule library, traversing the business rule library based on the file attribute data, and performing a first round of filtering; performing a second round of filtering based on the final scene label; merging the filtering results of the two rounds, removing conflicts, and generating feasible basic printing parameters to form a feasible basic printing parameter list; The content of calculating the utility value of each candidate scheme by the utility function includes: calculating the preference matching degree score, the cost score, the timeliness score, and the scene adaptation degree score of each candidate scheme respectively, and calculating the utility value of each candidate scheme based on the utility function. 6.The method of claim 5, wherein, The content of traversing the business rule library based on the file attribute data and performing a first round of filtering includes: Performing a second round of filtering based on the final scene label; The content of merging the filtering results of the two rounds, removing conflicts, and generating feasible basic printing parameters to form a feasible basic printing parameter list includes: Traverse the business rule library, match the file attribute conditions in the applicable conditions, extract and summarize the parameter constraints corresponding to the matching rules, and form a first round of filtering result set; wherein the business rule library contains rule ID, applicable condition, parameter constraint, and priority, and the parameter constraint includes mandatory parameter, recommended parameter, and prohibited parameter; Based on the final scene label, traverse the business rule library, match the scene conditions in the applicable conditions; extract and summarize the parameter constraints corresponding to the matching rules, and form a second round of filtering result set; Merge the parameter constraints of the two rounds of filtering result sets, retain the mandatory parameters and recommended parameters, and eliminate the prohibited parameters to form an initial merged set; Identify the conflict type, the conflict type includes numerical conflict, range conflict and logical conflict, and eliminate the conflict based on the priority; Integrate the eliminated parameter constraints, classify and arrange them according to paper type, page size, color mode, single or double side, binding method, and delivery time to obtain a feasible basic printing parameter list. 7.The method of claim 5, wherein, The specific steps of selecting a candidate scheme include: The M sets of initial schemes are encoded into feature vectors, the Euclidean distances between the initial schemes are calculated as the difference degrees, and a scheme difference degree matrix of MxM is constructed; the number of clustering groups K is determined according to the number of initial schemes; the M sets of schemes are divided into K clusters by using the K-means algorithm; for each cluster, the L schemes closest to the cluster center are selected as representative schemes; first, one representative scheme is selected from each cluster to form an initial candidate scheme set; the distribution of the initial candidate scheme set in the key dimensions, including delivery time and binding method, is checked, and based on the distribution, representative schemes are supplemented from the corresponding clusters; iteration is performed until N sets of candidate schemes are generated. 8.The method of claim 5, wherein, The content of selecting M sets of initial schemes from the feasible basic printing parameters in combination with the improved Monte Carlo sampling method includes: randomly selecting a value for each parameter category from the feasible basic printing parameters, but the selection probability is proportional to the assigned weight, and repeating the sampling until M sets of initial schemes are generated. 9.The method of claim 1, wherein, The content of generating presentation content based on the recommended scheme and the decomposition result of the utility function includes: S5.1, generating a recommended scheme description based on the printing parameters of the recommended scheme; S5.2, generating a recommended reason based on the scores in the utility function to obtain a recommended reason based on the utility function; S5.3, matching a display image based on the printing parameters of the recommended scheme, and the visual display content includes the display image. 10.The method of claim 1, wherein, The content of receiving user feedback information on the presentation content and generating a revised recommended scheme and current printing order information based on the feedback information includes: S6.1, obtaining display feedback of the user, including scheme confirmation and parameter customization modification; S6.2, obtaining implicit feedback of the user, including presentation content dwell time and abandoned order behavior; S6.3, when there is a user click on the scheme confirmation in the display feedback, automatically generating structured current printing order information based on the user-selected scheme; S6.4, when there is no scheme confirmation in the display feedback, revising the recommended scheme and the presentation content based on the display feedback and the implicit feedback, and pushing again to the user until a preset stop pushing condition is met.

Images