Marketing content dynamic updating method based on customer interest drift detection

Abstract

The present application relates to a marketing content dynamic updating method based on customer interest drift detection, obtaining behavior data of target customers and interaction process data of current marketing content; detecting whether the target customers have interest drift based on the behavior data; when the interest drift is detected, locating the mismatch position of the current marketing content based on the interaction process data; the current marketing content includes multiple independently updatable content layers; determining the corresponding target update layer according to the mismatch position; only performing update processing on the target update layer to generate updated marketing content, and keeping the content part not corresponding to the mismatch position unchanged; outputting the updated marketing content for putting to the target customers. By analyzing customer behavior and marketing interaction data, accurately identifying interest drift and locating mismatch position, targetedly correcting the corresponding content layer, improving marketing pertinence and conversion efficiency, and reducing resource consumption.

Description

Technical Field

[0001] This invention relates to the field of computer information processing technology, specifically a method for dynamically updating marketing content based on customer interest drift detection. Background Technology

[0002] Current technologies, such as the one published in CN118691350A (a method for managing advertising value based on intelligent targeting), collect user behavior and content feature data in real time. They reflect users' latest interests and behavioral patterns by dynamically updating user profile models and user interest fluctuation models. High-value content features are selected and the user profile model is updated using similarity, interest change rate, semantic relevance, user intent, and interaction rate indicators. Finally, relevant ads are matched from the advertising database. However, this approach still has significant limitations and structural defects. The core of this solution still revolves around the completeness of user profiles, the value of content features, and ad matching results. Its focus is on supplementing user profiles and reselecting ads, rather than directly addressing the interaction failure points of the current marketing content itself. Therefore, although it can know that user interests have changed and which content features are worth including in the model, it is difficult to further determine in which specific content part, which interaction node, or which type of guidance position the currently being delivered is mismatched with the customer's current interests. As a result, it cannot establish a closed-loop mechanism from interest drift to content mismatch to partial update. This leads to the actual system usually only being able to adopt coarse-grained update methods such as re-recommendation, overall replacement, or re-matching of ads, rather than fine-tuning a single mismatch layer.

[0003] While this document emphasizes using time series analysis to capture the changing trends and periodic fluctuations of user interests, and mentions that interest fluctuation models can reflect short-term and long-term interest changes, it essentially still focuses on numerical assessment of interest changes. It does not provide a mechanism for determining the order of visits to original interest topics and candidate new interest topics, nor does it establish constraints that better reflect true interest shifts, such as the first valid visited topic, the sequential substitution relationship across multiple consecutive behavioral cycles, and re-entry behavior. Therefore, in practical applications, it is still easy to misidentify short-term hotspot clicks, occasional visits, temporary browsing, or noisy behavior as interest changes, thus increasing the probability of false triggers in subsequent updates and reducing the stability and reliability of interest drift identification. While content feature supplementation can be triggered when the user profile completeness index falls below a preset threshold, and the impact of accidental fluctuations can be reduced by smoothing with moving averages and medians, this triggering logic still judges whether the profile quality is sufficient, rather than whether the customer has already been interrupted or jumped to a specific content segment, functional module, or guiding position in the current marketing process. In other words, it solves the problem of incomplete profiles, not the problem of mismatch in the internal structure of marketing content. Therefore, even if the profile is updated, the original path blockages, guidance failures, expression deviations, and hierarchical misalignments in the current marketing content may still be completely retained, and the root causes of customers jumping out, interrupting, or migrating across content during the interaction process cannot be truly eliminated.

[0004] This solution primarily utilizes high-value content features selected through screening, incorporating them into real-time behavioral data. This data is then used to further update the user profile model, followed by matching the most relevant ads from the advertising database. Essentially, this approach still updates the model first and then selects ads, failing to break down the current marketing content into multiple independently updatable layers. It also lacks a mechanism to map a target update layer based on mismatch locations and only update that target layer. Therefore, its update granularity is typically coarse. Responding to changes in interest often means recalculating profiles, rearranging creatives, rematching ads, or even replacing the entire campaign content. This not only increases the system's computational burden and resource consumption but may also render previously effective content that could still encourage customer interaction ineffective. This leads to a waste of existing effective content assets and may even disrupt the continuity of established conversion paths. Although the document mentions verifying the accuracy and effectiveness of the updated user profile model and interest fluctuation model through cross-validation or A / B testing, the object of this verification is still the model itself, rather than testing and updating the target update layer around the retained content. It also does not confirm the effectiveness of the partial update based on whether customers form a continuous preset response to the test update content. Therefore, its verification level is biased towards the effectiveness of the algorithm rather than the effectiveness of the content structure. It is difficult to guarantee that the updated marketing content can continue the customer's existing focus and fix the original mismatch in the real interaction path. As a result, at the business level, there may still be situations where the model is evaluated well but the actual conversion of the content is poor. Summary of the Invention

[0005] The purpose of this invention is to provide a method for dynamically updating marketing content based on customer interest drift detection, thereby addressing some of the drawbacks and shortcomings pointed out in the background art.

[0006] The present invention addresses the aforementioned technical problems by employing the following technical solution: a marketing content dynamic update method based on customer interest drift detection, comprising: acquiring target customer behavior data and interaction process data with current marketing content; and detecting whether the target customer has experienced interest drift based on the behavior data.

[0007] When the interest drift is detected, the mismatch position of the current marketing content is located based on the interaction process data;

[0008] The current marketing content includes multiple independently updatable content layers; the corresponding target update layer is determined based on the mismatch position; only the target update layer is updated to generate the updated marketing content, while keeping the content portion that does not correspond to the mismatch position unchanged; the updated marketing content is output for delivery to target customers.

[0009] Furthermore, the order in which the target customer visits the original interest topic and the candidate new interest topic is compared; when the candidate new interest topic changes from a non-first valid access topic to the first valid access topic in multiple consecutive behavioral cycles, and the original interest topic loses its status as the first valid access topic, it is determined that the target customer has experienced interest drift.

[0010] Furthermore, the interruption points in the interaction process and the jump landing points after the interruption are determined; when the interruption points continuously correspond to the same content part of the current marketing content, and the jump landing points correspond to the same functional content part in other marketing content, the content part is determined as a candidate mismatch position.

[0011] Furthermore, a portion of the content that can still trigger continued interaction from the target customer is identified from the content portion that did not correspond to the mismatched position, and this portion of the retained content is kept unchanged; the target update layer is tested and updated around the retained content portion; when the target customer generates a continuous preset response to the test update content, the test update content is output as the updated marketing content.

[0012] Furthermore, the first valid access topic of the target customer is recorded in units of preset behavior cycles; wherein, the valid access topic is the access topic in which the dwell time reaches a preset threshold or a preset interactive behavior occurs.

[0013] Furthermore, when the candidate new interest topic is accessed before the original interest topic in multiple consecutive behavior cycles, and the original interest topic is accessed later or not at all in the corresponding behavior cycle, the candidate new interest topic is determined to replace the original interest topic as the first valid access topic.

[0014] Furthermore, after the candidate new interest topic replaces the original interest topic as the first valid access topic, it is detected whether the target customer forms a re-entry behavior around the candidate new interest topic within the same behavior cycle; when the re-entry behavior is formed, the interest drift is further confirmed.

[0015] Furthermore, the content section where the jump landing point is located is compared with the content function of the content section; when both are used for the same guidance target, it is determined that the jump landing point corresponds to the content section with the same function.

[0016] Furthermore, after determining that the jump landing point corresponds to the same functional content section, it is detected whether the target customer continues to have a preset subsequent interaction after the same functional content section; when the preset subsequent interaction is successful, the candidate mismatch position is confirmed as the mismatch position.

[0017] Furthermore, after the preset subsequent interaction is completed, it is detected whether the target customer returns to the content section corresponding to the candidate mismatch position; when the target customer does not return to the content section and completes the subsequent interaction along the interaction path corresponding to the content section with the same function, the mismatch position is determined as a stable mismatch position.

[0018] The beneficial effects of this invention are as follows: By acquiring behavioral data of target customers and interaction data of current marketing content, changes in customer interests are dynamically identified. Upon detecting interest drift, the mismatch in the current marketing content is further located. This allows adjustments to marketing content to move beyond simple replacement or static recommendations, enabling targeted correction of specific mismatches corresponding to changes in customer interests. Compared to existing technologies that rely solely on historical tags or single clicks for coarse content adjustments, this invention compares the access order of original and candidate new interest topics over a continuous behavioral cycle. By combining the first valid access topic and re-entry behavior to confirm interest drift, it can more accurately identify the true trend of customer interest migration, improving the stability and reliability of interest judgment and avoiding misjudgments caused by short-term abnormal access.

[0019] After identifying interest drift, instead of completely reconstructing all marketing content, the process accurately pinpoints the mismatch by identifying breakpoints, jump points, and functional content sections. Based on these mismatches, a target update layer is determined, and updates are performed only on the corresponding content layer. Meanwhile, content sections that remain relevant and continue to engage customers are preserved, thus achieving localized dynamic optimization while maintaining the overall effectiveness of the marketing content. Through testing updates and validating them with continuous preset responses, updates that better match customers' current interests can be selected, improving the targeting, continuity, and conversion efficiency of marketing content delivery, reducing resource consumption from ineffective modifications, and enhancing the refinement and intelligence of the marketing content update process. Attached Figure Description

[0020] Figure 1 This is a logic diagram of the dynamic update of marketing content based on interest drift and interaction path in this invention; Figure 2 This is a heatmap of interest topic migration and interaction under continuous behavior cycles in Embodiment 1 of the present invention; Figure 3 This is a ladder diagram of the interest drift identification and local update triggering state in Embodiment 1 of the present invention; Figure 4 This is a trend graph of interest drift confirmation coefficient and smooth substitution intensity in Embodiment 1 of the present invention; Figure 5 This is a matrix diagram for determining the stable mismatch at the interruption position in Embodiment 2 of the present invention; Figure 6This is a polar coordinate mapping of the activation values ​​of the retained content fragments in Embodiment 2 of the present invention; Figure 7 This is a scan diagram of the test update benefit coefficient enhancement parameters in Embodiment 2 of the present invention. Detailed Implementation

[0021] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0022] Combined with appendix Figure 1 This invention relates to a dynamic marketing content update method based on customer interest drift detection. The system first determines the collection period corresponding to the target customer and continuously acquires the target customer's behavioral data and interaction data with the current marketing content within that period. Behavioral data may include access topics, access order, page dwell time, click behavior, favorites behavior, inquiry behavior, exit behavior, and re-entry behavior. Interaction data may include the target customer's browsing path after entering the current marketing content, dwell time in each content section, trigger operation sequence, interruption point, and jump destination. The system performs time alignment, noise reduction, and topic classification on the collected data to form a periodic behavioral sequence corresponding to the target customer.

[0023] The system extracts valid access topics within each behavioral cycle and compares the access order of the original interest topic and candidate new interest topics across multiple consecutive behavioral cycles. When a candidate new interest topic consistently appears earlier across multiple consecutive behavioral cycles, and generates more frequent dwell, click, or re-entry behaviors around that topic, while the access priority of the original interest topic decreases, dwell time shortens, or it is no longer accessed, the system determines that the target customer has experienced interest drift and outputs the interest drift detection results for use in subsequent marketing content updates.

[0024] Upon detecting a shift in target customer interest, the system retrieves interaction data from the corresponding behavioral period, analyzes the target customer's browsing path within the current marketing content, and reconstructs the complete interaction trajectory in chronological order of access time. The system identifies interruptions that occur during browsing and determines the corresponding interruption locations, marking the content section corresponding to each interruption location as a candidate analysis node. The system continues to track the target customer's redirection path after the interruption, obtains the content section corresponding to the redirection point, and analyzes the functional guidance objectives of that content section.

[0025] The system compares and analyzes the content corresponding to the interruption point with the content corresponding to the jump landing point. When both have consistent functional goals and are used to achieve the same guidance purpose, the system determines that the target customer did not complete the corresponding guidance in the current marketing content, but instead completed the guidance by jumping to other marketing content. Therefore, the content corresponding to the interruption point is identified as a candidate mismatch position. The system detects whether the target customer continues to generate preset interactive behaviors after the jump landing point and determines whether the interaction path is completed continuously. If the interaction is continuous and the customer does not return to the original interruption point, the candidate mismatch position is identified as a mismatch position in the current marketing content for subsequent update processing.

[0026] The system performs a structured analysis of the current marketing content, dividing it into multiple independently updatable content layers. Each content layer carries a different expressive function, including a display layer to attract attention, an explanatory layer to convey key information, and an interactive layer to guide actions. Based on the mapping relationship between the mismatch position and the overall content structure, the system determines the target update layer corresponding to the mismatch position and uses this target update layer as the sole object for subsequent adjustments.

[0027] During the update process, the system keeps other content sections in the mismatched areas unchanged to ensure that existing effective content that can continuously generate interaction from the target customers is preserved. For the target update layer, the system combines the detected interest drift results with the target customer's current interests and preferences to replace or reorganize the theme expression, information presentation method, or guidance path in that content layer, thereby generating updated content that matches the target customer's current interests. After the update is completed, the system recombines each content layer to form the updated complete marketing content and outputs the updated marketing content to the delivery module for targeted display to the target customers, achieving dynamic correction of local mismatched content without disrupting the overall structure.

[0028] After acquiring the target customer's behavioral data, the system divides the target customer's access records into preset behavioral cycles and sorts the access topics within each cycle. The system identifies valid access topics based on dwell time and interaction behavior, determining the first valid access topic in each cycle according to access time, while also recording the access position of the original interest topics in each cycle. The system extracts the changes in the access order of candidate new interest topics from the behavioral cycle sequence and tracks and analyzes their positions across multiple consecutive behavioral cycles.

[0029] When a candidate new interest topic is not in the first valid access position in the initial stage, but gradually advances and eventually stabilizes as the first valid access topic in subsequent consecutive behavior cycles, while the original interest topic no longer occupies the first valid access position in the corresponding behavior cycle, and its access order is significantly shifted or missing, the system determines that the candidate new interest topic has replaced the original interest topic as the dominant interest direction of the target customer, and confirms that the target customer has experienced interest drift, thus providing a basis for judgment for subsequent content updates.

[0030] After acquiring the target customer's behavioral data, the system segments continuous access records according to a preset behavioral cycle, and identifies and categorizes the target customer's access topics within each cycle. The behavioral cycle can be divided based on access time intervals or interaction continuity to ensure relative integrity of the behavior within each cycle. The system extracts all access topics within each behavioral cycle and determines the validity of the access based on dwell time and interaction behavior.

[0031] The system classifies access topics whose dwell time reaches a preset threshold as valid access topics, and also classifies access topics that involve clicks, favorites, inquiries, or other preset interactive behaviors as valid access topics. For access topics that meet both the dwell time and interactive behavior criteria, the system prioritizes retaining them as high-confidence valid access topics. After completing the screening of valid access topics, the system sorts the valid access topics according to the access time and determines the first valid access topic that appears within the behavior period as the first valid access topic.

[0032] After recording the first valid access topic within each behavioral cycle, the system compares and analyzes the access order of the original interest topics and candidate new interest topics across multiple consecutive behavioral cycles. The system constructs a topic access sequence according to the time order of the behavioral cycles and extracts the appearance position and chronological relationship between candidate new interest topics and original interest topics in each behavioral cycle.

[0033] During the analysis, the system determines whether candidate new interest topics are consistently accessed before the original interest topics across multiple consecutive behavioral cycles, and tracks the changes in the access order of the original interest topics within the corresponding behavioral cycles. When a candidate new interest topic consistently maintains a priority access position across multiple consecutive behavioral cycles, and the access order of the original interest topic is significantly delayed, or it does not appear in some behavioral cycles, the system considers the candidate new interest topic to have established a sustained advantage in access priority. Combining the valid access determination results within each behavioral cycle, the system confirms that the candidate new interest topic meets the valid access conditions in the corresponding behavioral cycle and consistently appears with priority across multiple consecutive behavioral cycles. Based on the above continuity and stability judgments, the system determines that the candidate new interest topic replaces the original interest topic as the first valid access topic, thus providing a basis for further confirmation of interest drift.

[0034] After identifying a candidate new interest topic as the first valid access topic to replace the original interest topic, the system further refines the access path analysis within the corresponding behavioral cycle to verify whether the topic has formed a stable interest concentration. Within the same behavioral cycle, the system tracks the target customer's access trajectory around the candidate new interest topic, identifies whether the content corresponding to that topic has been accessed multiple times, and records the time interval and interaction behavior of each access. Re-entry behavior includes the target customer's behavior of re-entering the same topic-related content after leaving the content corresponding to that topic through returning, clicking again, or other paths.

[0035] During the detection process, the system confirms that the target customer has made at least one valid initial visit to a candidate new interest topic within the same behavioral cycle. It then determines whether the customer re-enters the corresponding content within that topic during subsequent interactions. When the system detects that the target customer repeatedly enters a candidate new interest topic within the same behavioral cycle, with each entry accompanied by a pause or interaction, the system determines that the topic has generated sustained interest within the current behavioral cycle. Based on the interest reinforcement characteristics reflected by the re-entry behavior, the system further confirms that the target customer's interest has shifted from the original interest topic to the candidate new interest topic, thus completing the reinforcement determination of interest drift.

[0036] After parsing the interaction process data, the system reconstructs the browsing path of the target customer according to the access time sequence and identifies interaction interruption behaviors within this path. The system identifies the location where the target customer stops browsing the current marketing content and fails to complete the preset guidance goal as the interruption point. Simultaneously, it tracks the access path after this interruption, obtains the corresponding jump landing point, and records the content source and functional attributes of the jump landing point. The system performs statistical analysis on the interruption points over multiple behavior cycles. When the interruption point stably corresponds to the same content section in the current marketing content during continuous interaction, and the jump landing point corresponds to a content section in other marketing content used to achieve the same guidance goal, the system determines that the target customer has exhibited continuous deviation behavior in that content section, thus identifying that content section as a candidate mismatch location.

[0037] After identifying candidate mismatch locations, the system further analyzes other content sections in the current marketing content that do not correspond to these locations. It identifies content that can still prompt target customers to continue browsing, clicking, or performing preset actions, and designates this section as retained content. The system maintains the retained content unchanged during subsequent updates to preserve the stability of existing effective interaction paths. Around the retained content, the system only performs test updates on the target update layer corresponding to the candidate mismatch locations. Test updates include replacing or adjusting the theme expression, information presentation, or guidance methods, and generating test update content. The system delivers the test update content to target customers and monitors their response behavior in real time. When it detects that target customers continuously generate preset responses to the test update content and form a stable interaction path, the system confirms the test update content is effective and outputs it as updated marketing content for continued subsequent delivery.

[0038] After obtaining the breakpoint and its corresponding redirection point, the system performs functional attribute analysis on the content sections where the breakpoint and redirection point are located. The system extracts descriptive information about the role of each content section within the marketing content and, combined with the types of interactive behaviors it triggers, identifies the guidance objectives corresponding to each content section. Guidance objectives can be reflected as the intention to guide target customers to perform specific actions, such as entering a details page, initiating an inquiry, or completing key steps in the conversion path.

[0039] After identifying the guidance target, the system compares and analyzes the content corresponding to the interruption point with the content corresponding to the jump landing point to determine whether they point to the same guidance target. If the system identifies that the content at the jump landing point plays the same guiding role as the content at the interruption point and is used to achieve the same interactive goal, then the system determines that the jump landing point corresponds to the content with the same function.

[0040] After determining that the redirect point corresponds to the same functional content section, the system continues to track and analyze the target customer's interactive behavior after that section. Using the redirect point as the starting node, the system records the target customer's behavioral sequence in the subsequent access path and extracts behavioral features related to the preset interaction goals. These preset subsequent interactions include continuing to browse related content, performing click operations, entering the next guided page, or completing key steps in the target conversion path.

[0041] During the analysis, the system determines whether the target customer engages in continuous and effective interaction after the content section with the same function, and assesses whether the interaction path is complete and uninterrupted. If the system detects that the target customer continuously engages in preset subsequent interactions after the content section with the same function, and the relevant behaviors meet preset validity conditions, it indicates that the target customer has completed the guidance objective that should have been achieved in the current marketing content through the jump path. The system confirms that the previously marked candidate mismatch positions did not effectively carry the corresponding guidance function, thus determining the candidate mismatch positions as mismatch positions in the current marketing content, providing an accurate positioning basis for subsequent content updates.

[0042] The system determines whether a target customer exhibits a return behavior after completing a preset follow-up interaction. Return behavior includes re-entering the content section corresponding to the candidate mismatch location via a back button, a subsequent click, or a redirect. When the system detects that the target customer does not exhibit a return behavior during subsequent interactions and continues to complete subsequent access or conversion steps along the interaction path corresponding to the same functional content section, the system considers the target customer to have completely deviated from the original content path and formed a new interaction path centered on the same functional content section. Based on the stability characteristics of this path, the system identifies the mismatch location as a stable mismatch location, indicating that this content section has consistently lost its effective guiding role in the current marketing content, providing a reliable basis for subsequent targeted updates.

[0043] Example 1:

[0044] A platform continuously delivers a set of marketing content built around a pre-existing interest theme to target customers, guiding them to learn about a certain service plan and complete subsequent conversion interactions. The system collects data on the order of target customers' access to themes, dwell time, and interaction processes such as swiping, clicking, and re-entry over multiple consecutive behavioral cycles. Initially, target customers typically prioritize content corresponding to their pre-existing interest theme and maintain effective dwell time within that theme. As interactions continue, the system detects that candidate new interest themes are gradually appearing in the order of access, eventually replacing the pre-existing interest theme as the first effective access theme over multiple consecutive behavioral cycles, while also generating re-entry behavior around these candidate new interest themes. Based on this, the system determines whether target customers have experienced interest drift and triggers a partial dynamic update of the marketing content upon confirmation.

[0045] The current marketing content consists of multiple independently updatable content layers, including at least a cover guide layer, an interest expression layer, a benefits display layer, and a conversion action layer. The cover guide layer attracts target customers to the main content path; the interest expression layer presents the core solution around the target customers' current interests; the benefits display layer showcases benefits related to the theme; and the conversion action layer guides target customers to perform interactive actions such as claiming, inquiring, or further comparing. In subsequent updates, the system only replaces the interest expression layer, while the cover guide layer, benefits display layer, and conversion action layer remain unchanged, thus meeting the requirement of updating only the target update layer without altering other content layers.

[0046] The system records the first valid access topic in a unified preset behavior cycle. The criteria for a valid access topic are a dwell time of no less than 8 seconds, or the occurrence of clicks on details, expansion of descriptions, or addition of preset comparison interactions. The target customer's behavior data within 8 consecutive behavior cycles is shown in the table below, where a smaller access order value indicates an earlier access, and no access is recorded as 4.

[0047]

[0048] The system determines whether candidate new interest topics exhibit a substitution trend based on the first valid access topic in each behavioral cycle. As shown in Table 1, in cycles 1 and 2, the original interest topic was consistently the first valid access topic. In cycle 3, the access order of the candidate new interest topic advanced to first place, while the original interest topic fell to second place. From cycle 4 to cycle 8, the candidate new interest topic maintained its first access order for five consecutive behavioral cycles, while the original interest topic's access order was delayed to third place or not accessed at all. Therefore, the system determines that the candidate new interest topic has consistently been accessed before the original interest topic in multiple consecutive behavioral cycles and has replaced the original interest topic as the first valid access topic. From cycle 4 to cycle 8, target customers exhibited re-entry behavior around the candidate new interest topic, indicating that this access was not accidental but rather formed a reinforced attention path around the candidate new interest topic within the same behavioral cycle, thus satisfying the conditions for further confirmation of interest drift. Figure 2 It can be further seen that the advantage value of the candidate new interest topic is significantly higher than that of the original interest topic, and it has been the first effective access topic since the 4th period. At the same time, the re-entry behavior appears continuously, indicating that the target customer's access focus has shifted from the original interest topic to the candidate new interest topic, which is consistent with the description of interest drift gradually forming and steadily increasing in this embodiment.

[0049] To quantify the leading advantage of the original interest topic and candidate new interest topics in a continuous behavioral cycle, the system constructs the advantage value of the first effective access topic as follows:

[0050] in, Indicates the topic In the The order of access within each behavioral cycle Indicates the first The weights for each behavioral cycle are determined. Since later behavioral cycles better reflect the latest changes in interest, the weight sequence is 0.8, 0.9, 1.0, 1.1, 1.2, 1.3, 1.4, and 1.5. The candidate new interest topics are:

[0051] The original topics of interest include:

[0052] Therefore, the advantage value of the candidate new interest topics is significantly higher than that of the original interest topics, indicating that the candidate new interest topics have achieved a dominant position in the overall access priority.

[0053] The system further constructs an interest drift confirmation coefficient to simultaneously reflect both the continuous periodic substitution effect and the re-entry enhancement effect:

[0054] in, This indicates the number of re-entries formed around candidate new interest topics. Let's take 0.25. The table shows that the number of re-entries is 5. The weighted differences for the first 8 periods are approximately -0.2545, -0.2093, 0.2325, 0.3499, 0.3816, 0.4545, 0.4905, and 0.4771, respectively, with a sum of approximately 1.9223. Therefore:

[0055] And because:

[0056] Therefore, the interest drift confirmation coefficient is:

[0057] The system's preset interest drift confirmation threshold is 45%. Because... =68.82% is higher than 45%, therefore it can be confirmed that the target customers no longer form a stable priority access pattern around the original interest topic, but rather form a continuous cycle of substitution and re-entry reinforcement interaction pattern around candidate new interest topics. Figure 3As can be seen, the system state enters the substitution signal appearance stage in the 3rd cycle, enters the re-entry reinforcement path in the 4th cycle, and enters the drift confirmation and state update stage from the 6th cycle onwards. This state ladder evolution shows that the system does not directly update content based on access changes in a single cycle, but first identifies candidate new interest topics for early access, then confirms their continuous substitution relationship, and then completes reinforcement verification by combining re-entry behavior. Finally, after the state stabilizes and improves, it triggers local dynamic updates, thus maintaining consistency with the interest drift determination path in this embodiment.

[0058] To avoid misjudgments caused by abnormal fluctuations in a single behavioral cycle, the system also smooths the substitution trend in subsequent stages, constructing the smoothed substitution strength as follows:

[0059] in, Let's take 0.6. Smoothing calculations begin in the third cycle, with an initial value of... And calculated based on the local drift amount from the 3rd to the 8th period. , , , , , Then we have:

[0060] The system presets a smooth substitution intensity threshold of 50%. Calculation results show that, starting from the 6th cycle, The fact that the threshold has been consistently exceeded indicates that the replacement of the original interest topic by the candidate new interest topic is not an instantaneous fluctuation, but rather a continuous and stable trend of interest migration. Figure 4 Furthermore, it is shown that since the 6th period, the smooth substitution intensity has been continuously higher than the preset threshold, indicating that the substitution of the original interest topic by the candidate new interest topic is not an occasional fluctuation, but a stable and strengthening trend formed under the combined effect of multiple consecutive first visits and re-entry behaviors. This can support the conclusion in this embodiment that interest drift has been established and can trigger local dynamic updates.

[0061] Based on the aforementioned continuous behavioral cycle data, the substitution relationship of the first effective access topic, the re-entry behavior formed around the candidate new interest topic, and the joint calculation results of the interest drift confirmation coefficient and the smoothing substitution intensity, the system confirms that the target customer's interest has drifted from the original interest topic to the candidate new interest topic. The candidate new interest topic maintains its position as the first effective access topic in five consecutive behavioral cycles, while the access order of the original interest topic is continuously delayed or not accessed in the corresponding cycles. At the same time, the target customer forms re-entry behavior around the candidate new interest topic in all five cycles, which is sufficient to prove that the target customer has established a new and stable attention path around the candidate new interest topic.

[0062] Example 2:

[0063] A platform continuously displays a set of marketing content to target customers, guiding them to select and convert to a certain service plan. This marketing content is organized in layers so that when a mismatch is detected in certain content, only the corresponding layer is updated. The system continuously collects data on page dwell time, scroll depth, clicks on details, switching comparison items, jumps to other marketing content, completion of claim actions, and whether the user returns to the original path during multiple visits to the current marketing content. The system further discovered that in multiple visits, the target customer's visit was interrupted at a specific section explaining a plan within the current marketing content, followed by a jump to another section within another marketing content. After this jump, the customer continued to complete the preset subsequent interactions without returning to the original interruption point. Based on this interaction pattern, the system identified the interruption point, jump landing point, content segments with the same function, candidate mismatch locations, and stable mismatch locations. Upon confirming a stable mismatch location, the system only updates the target update layer.

[0064] The current marketing content includes a value-guiding layer, a solution explanation layer, an evidence enhancement layer, an action trigger layer, and a supplementary prompt layer. The core segments within the value-guiding layer are denoted as... This section is used to showcase the overall value of the service plan and attract target customers to continue browsing. The explanatory snippets in the plan description layer are denoted as... The comparative explanatory passage is denoted as ,in It undertakes the core comparison guidance function. The trusted summary fragment in the evidence enhancement layer is denoted as... This is used to display existing usage effects, evaluation summaries, and credibility information. The execution entry segment in the action triggering layer is denoted as... This is used to guide target customers to take actions such as claiming, making an appointment, consulting, or continuing to compare. Comparison summary snippets in other marketing content are denoted as... The system will confirm this later. , and It can still reliably trigger continued interaction, and is therefore identified as the retained content section, while the solution description layer... It is identified as the specific mismatched content part in the target update layer.

[0065] The system continuously records 10 interactions between the target customer and the current marketing content, as shown in Table 2 below. The system pre-sets subsequent interactions to include at least two of the following: clicking on benefits details, comparing execution plans, entering the action trigger layer, or completing the claiming action.

[0066]

[0067] The system identifies the interruption point and subsequent jump location in each interaction based on the target customer's scrolling stop position, dwell time, click interruption logs, and jump source path within the current marketing content. As shown in Table 2 above, in 7 out of 10 interactions, the interruption occurred within a segment of the current marketing content. The connection was interrupted at one point, and six of those interruptions redirected directly to snippets of other marketing content. In comparison, , and There were only one interruption each time, and the corresponding jumps did not establish a stable continuation of the interaction. Therefore, the system initially judged the content fragments as... It has significant repetitive interruption characteristics and will It was identified as the primary jump landing point. Figure 5 The differences in interruption count, number of jumps within the same function, number of subsequent interactions completed, number of times no return was achieved, and candidate mismatch stability for each interrupt segment are presented in a matrix format. The corresponding indicators are 7, 6, 6, and 6, respectively, with a candidate mismatch stability of 0.6000, significantly higher than other content segments. This result indicates that the candidate mismatch is highlighted in the graph, further validating the finding. It exhibits significant characteristics of repeated interruptions and alternative jumps.

[0068] The system does not determine mismatch solely based on the redirection behavior itself, but further compares the redirection endpoints. From the original interruption location The system analyzes the guiding objectives, semantic tags, and subsequent action relationships of both within the current content path. A comparison reveals that... This is used to guide target customers through the comparison of solutions and trigger the next action layer. Both systems perform the function of compressing and comparing multiple options, highlighting the differences and benefits, and guiding the next step of selection. Both aim to achieve the same goal: reducing the comparison costs for target customers and encouraging them to make a choice. Based on this, the system determines... This is considered as a functional part of other marketing content, thus satisfying the functional judgment condition in candidate mismatch location identification.

[0069] The system first divides the content into segments. The reason for identifying it as a candidate mismatch position is as follows: The interruption points in multiple consecutive interactions, and the main jump landing point after the interruption. It was identified as a portion of other marketing content with the same function. Subsequently, the system continued to monitor the target customers' [content / activity]. Whether a pre-set follow-up interaction occurs afterward. The data shows that in 7 instances... During the interrupted interaction, there were 6 jumps to Furthermore, all six instances completed at least two pre-set follow-up interactions, including clicking on benefits details, comparing execution plans, entering the action trigger layer, and completing the claiming action. Therefore, the system confirmed... This is not a typical departure location, but a mismatch location with a clear alternative path. The system continues to monitor whether the target customer returns during the aforementioned six interactions. None of the results were returned, and all followed... The system will eventually complete the subsequent interactions along the corresponding path. It was determined to be a stable mismatch location.

[0070] The system filters out content that can still encourage continued interaction from other content fragments that did not correspond to the mismatched position, and retains this content. Analysis reveals that the core fragments in the value guidance layer... Accessed 10 times out of 10 views, with 8 of those visits successfully triggering a scroll-down or click to continue browsing. Trusted digest fragment in the evidence enhancement layer. Accessed 9 times out of 9 views, with 7 of those visits triggering further comparison or details clicks. (Claim entry fragment in the action trigger layer.) Of the 8 visits, 6 triggered subsequent actions. In contrast, the non-core segments in the supplementary hint layer, although accessed 6 times, only triggered 2 further interactions, contributing less. Based on these results, the system will... , as well as The selected content will be retained and kept unchanged, except for the content in the scheme description layer. This is identified as the part to be updated in the target update layer. Figure 6 The activation values ​​of each content segment and their relative relationship with the retention threshold are displayed using a polar coordinate bar chart. The activation value is 1.2875. The activation value is 1.0782. The activation value of the [specific element] is higher than the 0.80 threshold, while the activation value of the supplementary prompt fragment is 0.3662, which is lower than the threshold. This graphical result is consistent with the conclusion of the retained content selection, namely... , and The content that should be retained should remain unchanged, while the supplementary information layer should not be retained as the main basis.

[0071] The system constructs candidate mismatch stability values ​​to quantify whether a content segment constitutes a stable mismatch position. The expression is as follows:

[0072] in, Indicates content fragments Number of interruptions This indicates the number of times the program jumps to the same content section after an interruption. This indicates the number of preset subsequent interactions to complete after the redirect. Indicates the total number of interactions. For a fragment... ,have , , , Substituting, we get:

[0073] The system's preset candidate mismatch stability threshold is 45%. Because... A percentage higher than 45% indicates Not only are there frequent interruptions, but each interruption also results in a highly concentrated redirection to the same functional content section, followed by the completion of preset subsequent interactions. This can be confirmed as at least a mismatch position. Combined with the condition that it did not return to the original position after 7 jumps, the system further upgrades and confirms it as a stable mismatch position.

[0074] The system further constructs a reserved content activation value to identify content fragments that do not correspond to a mismatched position but can still trigger continued interaction. The expression is as follows:

[0075] in, Indicates content fragments The number of times that trigger further interaction. This indicates the number of times the content segment has been viewed. This represents the average effective interaction intensity. (This refers to the core segments of the value guidance layer.) ,have , , .but:

[0076] Enhancement layer fragments of evidence ,have , , .but:

[0077] Action trigger layer fragment ,have , , .but:

[0078] For the supplementary prompt layer fragment, there are , , .but:

[0079] The system's preset activation threshold for retained content is 0.80. Therefore, it can be concluded that... and All values ​​above the threshold should be retained unchanged, while the supplementary prompt layer values ​​below the threshold should not be used as the basis for key retention.

[0080] Finally, the system builds a test update benefit coefficient to verify the effect of updating only the target layer. The expression is as follows:

[0081] Where ΔU represents the continuous preset response improvement amount, ΔC represents the improvement amount of subsequent interaction completion rate, and ΔB represents the change in interruption rate. =0.4, =0.4, =0.2. After the test update, the continuous preset response rate increased from 40% to 68%, so ΔU = 28%. The subsequent interaction completion rate increased from 52% to 76%, so ΔC = 24%. The interruption rate decreased from 34% to 18%, so ΔB = -16% is calculated based on the improvement. Substituting, we get:

[0082]

[0083] The system's preset test update benefit threshold is 15%. Because... A percentage higher than 15% indicates that test updates implemented only for the target update layer have significant benefits. Figure 7 China-Israel parameters and For the scan dimension, in a fixed The diagram illustrates the variation range of the revenue coefficient G, the revenue threshold boundary, and the location of the value point in the example, all under the condition of G=0.2. The value point in the example is located in the high-revenue region above the revenue threshold of 15%, corresponding to a revenue coefficient of 24.0%. This indicates that under the current parameter configuration, updating only the target update layer can significantly improve the continuous preset response and subsequent interaction completion rate, while reducing the interruption rate. The results shown in the diagram are consistent with the revenue coefficient calculation results above, further proving that updating only the target update layer... Significant benefits can be obtained.

Claims

1. A method for dynamically updating marketing content based on customer interest drift detection, characterized in that... include: Acquire behavioral data of target customers and their interaction data with current marketing content; Based on the behavioral data, detect whether the target customer's interest drift has occurred; Upon detecting the interest drift, the mismatch position of the current marketing content is located based on the interaction process data, wherein locating the mismatch position of the current marketing content includes: Analyze the browsing path of target customers in the current marketing content and reconstruct the complete interaction trajectory according to the access time sequence; Identify the interruption behavior that occurs during the browsing process of the target customer and determine the corresponding interruption location; track the jump path of the target customer after the interruption behavior, obtain the content part corresponding to the jump landing point, and analyze the guiding goal of the content part corresponding to the jump landing point in terms of function; The content corresponding to the interruption position is compared and analyzed with the content corresponding to the jump landing point. When the two have the same functional goal and are used to achieve the same guiding purpose, the jump landing point is determined to be the same functional content in other marketing content, and the content corresponding to the interruption position is determined as a candidate mismatch position. Detect whether the target customer continues to generate the preset subsequent interaction after the jump landing point, and determine whether the corresponding interaction path is completed continuously; when the target customer continues to generate the preset subsequent interaction after the jump landing point, the corresponding interaction path is completed continuously and does not return to the interruption position, the candidate mismatch position is determined as the mismatch position in the current marketing content. The current marketing content includes multiple independently updatable content layers; the corresponding target update layer is determined based on the mismatch position; only the target update layer is updated to generate the updated marketing content, while keeping the content portion that does not correspond to the mismatch position unchanged; the updated marketing content is output for delivery to target customers.

2. The marketing content dynamic update method based on customer interest drift detection according to claim 1, characterized in that, The order in which the target customer visits the original interest topic and the candidate new interest topic is compared; when the candidate new interest topic changes from a non-first valid access topic to the first valid access topic in multiple consecutive behavioral cycles, and the original interest topic loses its status as the first valid access topic, it is determined that the target customer has experienced interest drift.

3. The marketing content dynamic update method based on customer interest drift detection according to claim 1, characterized in that, Statistical analysis was performed on the interruption positions over multiple behavioral cycles to determine that the interruption positions consistently corresponded to the same content portion in the current marketing content during continuous interaction.

4. The marketing content dynamic update method based on customer interest drift detection according to claim 1, characterized in that, Identify, from the content portions that do not correspond to the mismatched position, the reserved content portions that can still trigger continued interaction from the target customer, and keep the reserved content portions unchanged; The target update layer is tested and updated around the retained content; when the target customer generates a continuous preset response to the test update content, the test update content is output as the updated marketing content.

5. The marketing content dynamic update method based on customer interest drift detection according to claim 2, characterized in that, The first valid access topic of a target customer is recorded in units of a preset behavior cycle; wherein, the valid access topic is the access topic in which the dwell time reaches a preset threshold or a preset interactive behavior occurs.

6. The marketing content dynamic update method based on customer interest drift detection according to claim 5, characterized in that, When the candidate new interest topic is accessed before the original interest topic in multiple consecutive behavior cycles, and the original interest topic is accessed later or not at all in the corresponding behavior cycle, the candidate new interest topic is determined to replace the original interest topic as the first valid access topic.

7. The marketing content dynamic update method based on customer interest drift detection according to claim 6, characterized in that, After the candidate new interest topic replaces the original interest topic as the first valid access topic, it is detected whether the target customer forms a re-entry behavior around the candidate new interest topic within the same behavior cycle; when the re-entry behavior is formed, the interest drift is further confirmed.

8. The method for dynamically updating marketing content based on customer interest drift detection according to claim 1, characterized in that, After the preset subsequent interaction is completed, it is detected whether the target customer returns to the content part corresponding to the candidate mismatch position; when the target customer does not return to the content part and completes the subsequent interaction along the interaction path corresponding to the content part with the same function, the mismatch position is determined as a stable mismatch position.

Images