User behavior tracking method and system for graphic materials

Abstract

This invention provides a method and system for tracking user behavior in text and image content, comprising: splitting text and image content into basic element units and assigning unique identifiers, embedding lightweight collection scripts and binding listeners; collecting effective interaction behavior data on user terminals; preprocessing the data in the backend to obtain a single user access behavior sequence; storing the data using a distributed database architecture and building a multi-dimensional index; performing multi-dimensional analysis and calculation of element interaction popularity through models to mine user preferences and behavioral paths; visually outputting analysis reports and synchronizing interest data to downstream business systems. The system has six functional modules to implement the above method. This invention achieves element-level accurate full-link tracking, providing rich and efficient data processing, and can provide data support for content optimization, user profile construction, and precision marketing, while also protecting privacy and adapting to large-scale application scenarios.

Description

Technical Field

[0001] This invention relates to the field of computer data processing technology, and more specifically, to a method and system for tracking user behavior in graphic and textual materials. Background Technology

[0002] In the current online environment, analyzing user online behavior is a key technology for achieving personalized recommendations, optimizing content creation, and precision marketing. Existing technologies include several methods for analyzing user browsing behavior. These methods typically rely on macro-level behavioral data at the browser level, such as user dwell time, visit frequency, and bounce rate, to assess the user's overall interest in the page content. However, these methods are limited in that they treat a text / image resource as an indivisible whole, only tracking macro-level interaction data related to that whole, and failing to delve into the element-level interactions within the text / image resource itself.

[0003] Patent document CN101770520A discloses a user interest modeling method based on user browsing behavior, comprising two parts: explicit construction of the user interest model and implicit updating of the user interest model. Explicit construction of the user interest model is the initial establishment and initialization process of the user interest model through user registration. Implicit updating of the user interest model is the analysis and research of user browsing preferences based on the visitor's access to web pages without user participation.

[0004] In summary, existing technologies cannot accurately capture the differences in user interest in a specific text, image, or link within a content, resulting in insufficiently refined user profiles. This limits the guidance available to content creators for optimizing specific content elements and makes it difficult to meet the need for dynamic and refined tracking of user interests.

[0005] Therefore, the market needs a user behavior tracking method and system for graphic and text materials that can accurately and comprehensively track various behaviors during user interaction with graphic and text materials, obtain rich user behavior data, and provide data support for material optimization, user profile construction, and precision marketing. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for tracking user behavior in graphic and textual materials. This solves the technical problem that existing technologies can only perform macro-level behavior tracking on graphic and textual materials, and cannot delve into the material to obtain user interaction data on specific elements, thus resulting in insufficient accuracy in user interest analysis.

[0007] A method for tracking user behavior of text and image materials according to the present invention includes: Step S1: The graphic and text materials to be analyzed are structurally split into multiple basic element units, each basic element unit is assigned a unique identifier, a lightweight behavior collection script is embedded in the graphic and text materials, and an interactive event listener is bound to each basic element unit. Step S2: Monitor and collect effective interaction behavior data between the user and each basic element unit in real time on the user terminal through the behavior collection script; Step S3: The collected valid interaction behavior data is preprocessed on the backend server to obtain the behavior sequence of a single user access session; Step S4: The behavior sequence is stored using a distributed database architecture, and a multi-dimensional index is built based on the key fields of the behavior data to support fast query and statistical analysis; Step S5: Based on stored data and multidimensional indexes, perform multidimensional analysis through a preset statistical analysis model, including calculating the interaction heat score of each basic element unit and mining user interests, preferences and behavioral paths; Step S6: Output the analysis results in a user behavior analysis report in a visual form, and simultaneously synchronize the structured user interest data to the downstream business system to achieve data-driven business optimization.

[0008] Preferably, the behavior collection script includes a core collection layer responsible for implementing the core logic of data collection, an event listening layer responsible for binding and listening to user page interaction events, and a data reporting layer responsible for packaging the collected data and sending it asynchronously to the backend server.

[0009] Preferably, in step S2, the effective interactive behavior must meet the preset judgment criteria, including at least one of browsing, clicking, staying, collecting, sharing, and downloading. Valid browsing requires that the element enters the visible area and the duration reaches the element type-specific threshold. Valid clicking requires that a valid interactive event is triggered, the click duration exceeds the threshold, and the coordinates fall within the element's valid interactive area. For images and long text elements, we also collect in-depth behavioral data, including zooming, dragging, and reading progress.

[0010] Preferably, in step S5, the interaction popularity score of the basic element unit is calculated by weighting interaction frequency and interaction duration, and the calculation formula is as follows:

[0011] in, and All are weights. Values ​​0.6 The value is 0.4, or can be dynamically adjusted according to business needs.

[0012] Preferably, in step S5, the multi-dimensional analysis also includes an analysis of the overall interaction of the materials, specifically by counting the total number of views, the number of effective users, the average page dwell time, the bounce rate, and the overall interaction coverage of the elements. The user behavior path analysis uses a sequence pattern mining algorithm to reconstruct the typical interaction paths of users in graphic and text materials.

[0013] According to the present invention, a user behavior tracking system for graphic and textual materials is provided for implementing the method, comprising: a structured processing module, a data acquisition module, a data processing module, a data storage and indexing module, a statistical analysis module, and a result application module; The structured processing module is used to split the graphic and text materials into basic element units, assign unique identifiers, embed behavior collection scripts and bind event listeners, and output structured graphic and text materials; The data acquisition module is a lightweight behavior acquisition script embedded in structured graphic materials, which collects effective interactive behavior data on the user terminal and reports it to the backend. The data processing module is used to preprocess the raw behavioral data to form a user access behavior sequence; The data storage and indexing module uses a distributed database composite architecture to store behavioral data and constructs a multi-dimensional index. The statistical analysis module performs multi-dimensional statistical analysis through a preset model, calculating element interaction popularity and mining user preferences and behavioral paths. The results application module generates a visual report from the analysis results and synchronizes user interest data to downstream business systems.

[0014] Preferably, the structured processing module integrates a parser and an identifier generator; The parser traverses the structure of the source document to identify basic element units; The identifier generator generates unique identifier information for each unit according to preset structured coding rules.

[0015] Preferably, the data processing module has built-in filters, converters, and a session manager; The filter removes invalid and redundant behavioral data, the converter transforms all behavioral data into a unified standard data structure, and the session manager aggregates the behavioral sequence of a single visit based on the user's anonymous identifier.

[0016] Preferably, the data storage and indexing module adopts a combined architecture of HBase distributed columnar database and Elasticsearch distributed search engine; HBase is used for high-concurrency writing and persistent storage of massive behavioral data; Elasticsearch is used to synchronize key fields of behavioral data and build multidimensional indexes, supporting complex queries and aggregation analysis at the millisecond level.

[0017] Preferably, the result application module includes a visualization submodule and an interface service submodule; The visualization submodule renders the analysis results into a web dashboard / analysis report that includes element heatmaps, indicator trend charts, and user path maps; The interface service submodule encapsulates structured user interest and preference data into standard interfaces for downstream business systems to call, forming a data-driven closed loop.

[0018] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention achieves precise positioning and individual tracking of each basic element unit by structurally decomposing graphic and textual materials and assigning a unique identifier to each basic element unit. This breaks through the limitation of existing technologies that can only track overall macro data, and can obtain detailed behavioral data of users on different elements, providing a data foundation for accurate analysis of user interests.

[0019] 2. This invention achieves real-time and comprehensive collection of user behavior data by embedding lightweight behavior collection scripts into the materials, covering various behavior types such as browsing, clicking, dwelling, and favorites, ensuring the richness and completeness of the behavior data; combined with data preprocessing and multi-dimensional index construction, it improves the accuracy of the data and the efficiency of subsequent retrieval.

[0020] 3. This invention employs a multi-dimensional statistical analysis model, which can conduct in-depth behavioral analysis from multiple levels, including the overall material, element units, and individual users, and generate comprehensive user behavior analysis reports. This not only provides targeted optimization suggestions for material creators, but also provides accurate data support for personalized recommendation systems, effectively improving the dissemination effect and conversion efficiency of graphic and text materials.

[0021] 4. The overall solution provided by this invention has a clear process and strong operability. By using anonymous identification information to collect basic user data, it can achieve refined tracking while ensuring user privacy and security. The application of distributed database ensures the stability and efficiency of the system in processing large-scale behavioral data, making it suitable for user behavior tracking scenarios with various large-scale graphic and text materials. Attached Figure Description

[0022] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 A flowchart illustrating a method for tracking user behavior of graphic and textual materials provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a user behavior tracking system for graphic and textual materials provided in an embodiment of the present invention; Figure 3 This is a network topology diagram of a user behavior tracking system for graphic materials provided in an embodiment of the present invention. Detailed Implementation

[0023] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0024] Example 1 According to the present invention, a method for tracking user behavior of graphic and textual materials is provided, such as... Figure 1 As shown, it includes: Step S1: Structural Processing of Text and Image Materials. This step involves preprocessing the text and image materials to be analyzed, making them suitable for detailed tracking. Specifically, it may include the following operations: First, the target text and image material is structurally broken down. A typical text and image material, such as a blog post, product introduction page, or news report, is usually composed of multiple types of elements. This step identifies and divides multiple basic element units by parsing the document object model structure of the material's underlying code (e.g., Hypertext Markup Language). These basic element units are the smallest meaningful units that constitute the text and image content. In this embodiment, they may include, but are not limited to: titles, abstracts, body paragraphs, images, charts, tables, audio players, video players, clickable hyperlinks and function buttons (such as download and purchase), attachments, etc.

[0025] Subsequently, each basic element unit obtained from the split is assigned a unique identifier. This identifier is the core basis for locating specific elements in all subsequent data collection and analysis stages. To ensure the uniqueness, readability, and scalability of the identifiers, a structured encoding rule can be adopted. For example, it can be set to the format of: element type code + separator + element number. Specifically, the element type code can use predefined letters or abbreviations, such as T for title, P for paragraph, IMG for image, LNK for link, and TBL for table. The separator can be a hyphen "-" or an underscore "_". The element number is incremented from 0 or 1 according to the order in which the element of this type appears in the material. Accordingly, the first title in a piece of material can be identified as T-01, the third image can be identified as IMG-03, and the second link can be identified as LNK-02. After this processing, the original text and image material is transformed into a structured text and image material containing multiple basic element units with unique identifiers.

[0026] Furthermore, a behavior collection script is embedded in the processed structured text and image materials. This script is used to collect real-time interaction data between the user and each basic element unit, which is key to achieving refined data collection on the front end. To avoid affecting the user's page loading speed and experience, this behavior collection script needs to be lightweight. In this embodiment, the script can be written in native JavaScript without relying on any third-party libraries to ensure its versatility and execution efficiency. The overall code size of the script can be strictly controlled, for example, less than 5 kilobytes after compression and optimization. Architecturally, the script can be designed as a three-layer structure: a core collection layer responsible for implementing the core logic of data collection, an event listening layer responsible for binding and listening to user page interaction events, and a data reporting layer responsible for packaging the collected data and sending it asynchronously to the backend server. The embedding method is usually to embed the script file when generating the final page code. <script>标签引入。同时，在每个基础元素单元对应的标签上，通过编程方式添加相应的事件监听器。例如，为一个图片元素（标签）添加鼠标进入、鼠标离开、点击等事件的监听函数。所述监听函数与行为采集脚本中的逻辑相关联，当用户与页面元素发生交互并触发这些事件时，对应的监听函数即被调用，进而激活行为采集脚本以执行数据采集任务。

[0027] 步骤S2：用户行为数据采集。当用户通过浏览器或应用程序的内置网页视图访问加载了上述结构化图文素材的页面时，嵌入的行为采集脚本随页面一同加载并立即开始执行，实时监测并采集用户与页面上各个基础元素单元的交互行为数据。此步骤在用户终端（例如用户的个人电脑、手机或平板电脑）上完成。

[0028] 采集的数据是多维度的，以确保后续分析的全面性。在本实施例中，采集的行为数据至少包括：1.用户基础信息：为了在保护用户隐私的前提下区分不同用户并关联其行为，此处采集的是匿名的标识信息。例如，可通过读取浏览器的用户代理字符串，分析得出用户的设备类型、操作系统、浏览器类型等信息。在一些生态闭环的应用中，可通过调用平台提供的软件开发工具包接口获取一个针对该用户的、在当前应用内唯一的匿名身份标识符。这些信息有助于进行用户群体的划分和分析。

[0029] 2.交互行为类型及对应的基础元素单元标识：行为采集脚本需能够识别并记录多样的交互行为类型。每当一个行为被捕获时，脚本会同时记录该行为的类型以及发生该行为的基础元素单元的唯一标识信息（即在步骤S1中分配的标识）。

[0030] 在本发明的一个具体实现中，对交互行为类型的判定设定了明确且细化的标准，以提高数据准确性并排除无意义的噪音数据。例如：浏览：一次有效的浏览行为的判定，可综合考虑多个条件。其一，该元素必须进入到浏览器的当前可视区域内；其二，元素在可视区域内的持续时间必须达到一个预设的阈值，例如对文字类元素可设为3s，对图片类元素可设为1s，以过滤掉用户快速滚动页面时的一扫而过；其三，对于可滚动的长文本或大图，还可设置一个最低阅读或查看进度的要求，例如用户至少滚动查看了该元素30%的内容。

[0031] 点击：一次有效的点击行为的判定，需满足以下条件：捕获到鼠标左键单击或触摸屏的轻触事件；为排除用户无意识的误触，可设置一个最短点击时长（如100ms），时长过短的点击可能被视为无效；并且，点击的坐标位置必须落在元素的有效交互区域内。

[0032] 停留：指用户将注意力集中在某一元素上。一次有效的停留行为的判定，要求元素在交互期间全程保持在可视区域内，且用户在此期间没有进行其他显著操作（如大幅度滚动页面或点击其他元素），同时停留的时长需超过为该元素类型专属设定的阈值，例如标题为5s，一段核心正文为8s，一张复杂的图表为10s。

[0033] 收藏：判定用户收藏了某个元素，需捕获到用户点击了与该元素关联的收藏按钮或图标，并且收藏操作成功完成（例如，按钮状态从未收藏变为已收藏）。需要说明的是，脚本应能识别并排除用户紧接着又进行取消收藏的反向操作。

[0034] 分享：判定用户分享了某个元素，需捕获到用户触发了分享操作，并且成功调出了分享渠道选择界面或完成了分享动作。同样需要排除用户在分享过程中途取消的行为。

[0035] 下载：判定用户下载了与某元素关联的附件，需捕获到下载链接的点击事件，并确认浏览器的下载请求已成功发起。

[0036] 此外，针对特定类型的元素，行为采集脚本还可采集更深层次的行为数据。例如，对于图片类型的基础元素单元，可采集用户是否进行了放大、缩小、拖拽查看图片细节，以及鼠标指针在图片上停留的主要位置区域等数据。对于正文段落类型的基础元素单元，则可结合滚动事件和元素位置，动态计算用户的阅读进度，如已阅读字数、估算的阅读速度等。

[0037] 3.交互时间与交互时长：对于每一个捕获到的有效交互行为，脚本都会记录其发生的精确时间戳，以及行为的持续时长，如浏览时长、停留时长。

[0038] 步骤S3：行为数据预处理。用户终端的行为采集脚本将采集到的原始行为数据通过网络请求（例如，使用HTTP POST请求，将数据以JSON格式置于请求体中）传输至后端服务器的数据处理模块。由于前端采集的原始数据可能存在各种问题，直接入库分析会影响结果的准确性，因此需要进行预处理。此步骤在后端服务器上执行，主要包括：1.数据清洗：该操作旨在剔除无效和冗余的数据。例如，通过检查数据格式，丢弃字段不完整或格式错误的记录；通过检查用户标识和时间戳，识别并去除完全重复的数据记录。更重要的是，根据预设的业务规则过滤掉无意义的交互数据，例如交互时长远小于预设阈值的记录（如停留时间小于1s），或在极短时间内（如1s内）由同一用户对同一元素产生的多次重复行为记录。这些预设阈值可根据图文素材的整体类型和目标用户群的平均阅读习惯进行动态调整。

[0039] 2.数据标准化：由于数据可能来自不同版本的前端脚本，或在未来系统升级后产生新的数据格式，因此需要将所有接收到的行为数据转换为统一、标准的数据结构。例如，统一时间格式为协调世界时时间戳，统一将行为类型编码为标准的枚举值等。

[0040] 3.数据关联：将来自同一用户的、在不同时间点上报的、针对不同元素的多条离散行为数据，通过其共有的用户基础信息（匿名标识）进行关联，聚合形成一个能够反映该用户在一次完整访问会话中的行为序列。例如，用户A在访问某素材时的行为序列可能是：[浏览T-01, 浏览P-01, 浏览IMG-01, 停留IMG-01, 点击LNK-01]。这种行为序列对于分析用户行为路径至关重要。

[0041] 步骤S4：行为数据存储与索引构建。经过预处理后的用户行为数据需要被持久化存储，以便进行后续的查询和分析。考虑到用户行为数据具有产生频率高、数据量大的特点，本实施例采用分布式数据库方案以确保系统的可扩展性与高性能。

[0042] 具体地，可采用一种组合架构。核心的数据存储可选用HBase这类分布式、列式存储数据库，其非常适合海量数据的快速写入和基于行键的快速读取，能够很好地适配前端脚本高并发、批量上报数据的场景。存储结构可设计为以"用户ID-素材ID-会话ID”等组合作为行键，将一次会话中所有元素的所有行为数据存储在不同的列族和列中。例如，可设计一个"用户-素材-元素-行为”的四级逻辑结构，从而方便地查询某一用户对某一素材的所有交互，或某一元素被所有用户的交互情况。同时，为了支持复杂、多维度的即席查询和统计分析，可引入Elasticsearch这类分布式搜索引擎作为辅助。每当一条行为数据写入HBase后，通过数据同步机制（如消息队列或专用同步工具）将其中需要用于检索和聚合分析的关键字段（如用户基础信息中的标签、素材标识、元素标识、行为类型、交互时间等）同步至Elasticsearch中，并基于这些字段构建多维索引。Elasticsearch的全文检索和聚合分析能力，可为后续的统计分析提供毫秒级的查询响应，极大地提升了数据分析的效率和实时性。

[0043] 步骤S5：行为数据统计与分析。基于存储在分布式数据库中的用户行为数据以及构建好的多维索引，通过预设的统计分析模型对数据进行深入的多维度分析，最终生成具有洞察力的用户行为分析报告。

[0044] 统计分析模型的构建是一个系统性工程。首先，根据业务需求拆解分析目标，例如，评估素材整体吸引力、找出最受欢迎和最不受欢迎的元素、挖掘不同用户群体的兴趣偏好、还原主流的用户转化路径等。然后，为每个分析目标定义全维度可量化指标体系。接着，选择合适的算法（如加权评分、序列模式挖掘、聚类分析等）来计算这些指标，并搭建起一个"基础数据计算层-维度分析层-结果输出层”的三层模型架构。该模型会直接与HBase和Elasticsearch进行数据交互。

[0045] 在本实施例中，分析的维度可包括：1.素材整体交互情况：分析素材的总浏览量、有效浏览用户数、平均页面停留时长、跳出率，以及所有元素的总体交互覆盖率等宏观指标。2.各基础元素单元的交互热度：作为本方案的核心分析维度之一，可通过对每个元素的各项交互指标进行加权计算，得出一个综合的热度评分。可以理解的是，交互频次（如点击次数）反映了用户是否关注该元素的意愿，而交互时长（如停留总时长）则反映了用户对该元素投入的深度。可为这两者设置不同权重，例如0.6和0.4。其计算公式可表示为：

[0046] 通过计算每个基础元素单元的热度评分并进行排序，可以直观地识别出哪些内容最吸引用户，哪些内容被用户忽视。 3.用户兴趣偏好：通过聚合分析特定用户或用户群体与不同类型基础元素单元（如图片、视频、链接）的交互热度，可以描绘出其内容消费偏好。例如，若分析发现用户A对图片元素的交互热度远高于文字段落，则表明其可能是一个视觉导向型用户。 4.用户行为路径：通过分析在步骤S3中形成的用户行为序列数据，可以还原出用户在图文素材中的典型交互路径。例如，可能会发现大部分最终完成购买的用户都遵循了"查看标题 ->查看核心卖点图 ->查看用户评价段落 ->点击购买链接”这一路径，这对于优化转化漏斗非常有价值。

[0047] 步骤S6：分析结果输出与应用。最后，将步骤S5中生成的各种分析结果以直观、易懂的形式输出，并应用到实际业务中。

[0048] 一种应用方式是，将分析结果整合成一份详尽的用户行为分析报告，通过可视化的仪表盘或报表形式提供给素材的创作者或运营者。报告中可包含元素热力图、关键指标趋势图、用户路径图等，并附上基于数据得出的具体优化建议。

[0049] 另一种应用方式是，将分析得出的结构化用户兴趣偏好数据格式化后，通过接口实时或准实时地同步至个性化推荐系统、广告投放系统等下游应用中。这些精准的用户兴趣数据可以提升推荐内容与用户兴趣的匹配度，从而实现更精准的个性化推送，提高整体的用户参与度和转化率。

[0050] 实施例 2本发明还提供一种图文素材的用户行为追踪系统，所述图文素材的用户行为追踪系统可以通过执行所述图文素材的用户行为追踪方法的流程步骤予以实现，即本领域技术人员可以将所述图文素材的用户行为追踪方法理解为所述图文素材的用户行为追踪系统的优选实施方式。

[0051] 根据本发明提供的一种图文素材的用户行为追踪系统，如图2所示，该系统是实现实施例1所述方法的硬件与软件实体。该系统可部署在一台或多台服务器上，并通过网络与用户终端进行通信。

[0052] 如图2所示，该系统主要包括：结构化处理模块、数据采集模块、数据处理模块、数据存储与索引模块、统计分析模块以及结果应用模块。

[0053] 结构化处理模块，用于执行前述方法中的步骤S1。该模块接收原始图文素材作为输入，其内部集成有解析器和标识生成器。解析器负责遍历素材的文档结构以识别各个基础元素单元，标识生成器则根据预设的编码规则为每个单元生成唯一标识信息。最后，该模块将一个预置的行为采集脚本（即数据采集模块）的引用插入到素材的代码中，并为各个元素的标签绑定事件监听器，最终输出结构化的、可被追踪的图文素材。该模块通常在内容发布或更新时被调用。

[0054] 数据采集模块，在物理上并非后端模块，而是指被嵌入到用户所浏览页面中的轻量化行为采集脚本。如图2所示，直接与用户发生交互。当用户在浏览器中进行浏览、点击、滚动等操作时，该脚本被触发，负责实时捕获这些交互事件，并根据内置的判定逻辑（如时间阈值、可见性判断）筛选出有效行为。随后，该模块将包含用户匿名标识、元素标识、行为类型、时间戳等信息的行为数据打包，通过网络上报给后端的入口，即数据处理模块。

[0055] 数据处理模块，作为一个后端服务模块，是前端数据上报的接收端，其核心功能是执行前述方法中的步骤S3。该模块接收来自数据采集模块发送的原始行为数据流，并启动预处理流水线，该流水线依次执行数据清洗、数据标准化和数据关联操作。例如，会配置一个过滤器以丢弃交互时长小于1s的无效数据；使用转换器将所有传入的数据记录统一成内部标准的数据对象格式；并维护会话管理器，根据用户标识将属于同一次访问的行为数据聚合成行为序列。处理完成后的干净数据被传递给数据存储与索引模块。

[0056] 数据存储与索引模块，负责执行前述方法中的步骤S4。该模块内部管理着一个或多个分布式数据库系统。在本实施例中，可包含HBase集群用于主要的数据写入和存储，以及Elasticsearch集群用于索引和查询。该模块提供标准的数据库访问接口，供数据处理模块写入数据，并供统计分析模块读取数据。当新数据写入时，还负责触发数据同步机制，确保HBase中的数据能够准实时地在Elasticsearch中建立索引，从而保证数据的一致性与可检索性。

[0057] 统计分析模块，负责执行前述方法中的步骤S5。该模块内固化了多种统计分析模型和算法，可根据预设计划（例如，每小时执行一次）或由外部触发来启动分析任务。执行任务时，向数据存储与索引模块发起高效的查询请求（主要通过Elasticsearch的聚合查询接口）以获取所需的基础数据。然后，模块内的计算引擎会运行加权评分算法来计算元素热度，运行序列挖掘算法来发现用户行为路径，或运行聚类算法来划分用户群体。分析完成后，将生成结构化的分析结果，如JSON格式的报表数据或用户标签数据。

[0058] 结果应用模块，作为系统的输出端，负责执行前述方法中的步骤S6。该模块接收来自统计分析模块的分析结果，并将其转化为对不同下游友好的格式。例如，将报表数据渲染成一个可视化的Web仪表盘，供素材创作者访问和查阅，以帮助他们理解用户行为并获得优化建议。同时，也可将用户兴趣偏好数据封装成标准的接口，供个性化推荐系统或其他业务系统调用，从而将分析洞察转化为实际的业务动作，形成一个完整的数据驱动闭环。

[0059] 这六个模块协同工作，构成了一个从数据产生、采集、处理、存储、分析到应用的完整链路，实现了对图文素材用户行为的端到端精细化追踪与分析。

[0060] 实施例3为了更具体地说明本申请实施例提供的方法和系统在实际场景中的应用及其技术架构，本实施例将以一个电商平台的"便携咖啡机种草图文”为例，并结合系统的网络部署拓扑进行详细阐述。

[0061] 如图3所示，其为本发明实施例提供的一种图文素材的用户行为追踪系统的网络拓扑示意图。该图展示了系统各组件在网络环境中的典型部署方式。

[0062] 用户终端可以是用户的个人电脑或移动设备上的浏览器或应用程序。当用户希望访问该篇关于便携咖啡机的图文素材时，用户终端会发起一个HTTP请求。该请求通常不会直接到达源服务器，而是首先被内容分发网络拦截。内容分发网络在全球各地部署有缓存节点，从离用户最近的节点返回图文素材的静态内容（HTML、CSS、图片以及被嵌入的行为采集脚本），加快了页面的加载速度。

[0063] 用户终端加载完页面后，行为采集脚本（即数据采集模块）开始运行。当用户在页面上产生交互行为时，例如点击了一张展示咖啡机萃取过程的动图，脚本会捕获此事件，并封装一个包含元素标识IMG-04、行为类型Click等信息的数据包。随后，脚本通过一个异步的HTTP POST请求，将这个数据包上报至应用服务器。

[0064] 应用服务器是整个后端服务的核心，承载了本发明实施例中的结构化处理模块（在内容发布时使用）、数据处理模块、统计分析模块和结果应用模块。当应用服务器接收到上报的行为数据后，数据处理模块立即对其进行清洗、标准化和关联。

[0065] 处理后的数据需要被存储。应用服务器与后端的分布式数据存储集群进行交互。具体来说，将结构化的行为数据记录写入HBase集群。HBase集群由多个节点组成，保证了数据存储的高可用性与横向扩展能力。同时，为了支持快速查询，应用服务器会将数据的关键维度（如用户ID、元素ID、时间戳、地域等）同步到Elasticsearch集群中。Elasticsearch集群会对这些数据建立倒排索引，使得后续的统计分析模块能够以高性能执行复杂的聚合查询。

[0066] 下面以"便携咖啡机种草图文”的具体分析流程为例进行说明：素材与用户设定：素材：该图文素材经过结构化处理模块处理后，被拆分为12个基础元素单元，并分配了唯一标识。例如：标题为T-01，一段介绍核心卖点的正文为P-03，一张展示产品操作步骤的图片为P-02，一个指向商品购买页的主要链接为L-01，一个包含详细技术参数的表格为TBL-01。

[0067] 用户：假设在一天内有1000名获取了匿名授权的用户访问了该页面。

[0068] 系统配置：行为采集脚本体积为4.8KB；交互热度评分模型中，交互频次的权重为0.6，交互时长的权重为0.4；无效交互的过滤阈值为交互时长小于1s。

[0069] 执行流程与数据分析：数据采集与预处理：在一天内，系统共采集到10373条原始交互记录。数据处理模块启动预处理流程，其中123条记录因交互时长过短或为重复数据而被剔除，无效数据占比约为1.2%。剩余的10250条有效数据被存入HBase集群并同步至Elasticsearch集群。

[0070] 统计分析：预设的分析任务启动，统计分析模块向Elasticsearch集群发起查询。

[0071] 元素热度分析：模块计算出每个基础元素单元的交互热度评分。结果发现，热度排名前三的分别是：操作步骤图片P-02（热度评分92.4），主要购买链接L-01（热度评分88.6），以及一段讲述新手易犯错误及避坑指南的正文P-05（热度评分76.2）。而参数表格TBL-01的热度评分仅为18.5，排名垫底。

[0072] 用户偏好分析：通过对所有用户交互行为的类型进行聚合，发现图片类元素的总交互热度占比达到42%，链接类为28%，正文类为20%。这表明对于这款产品，用户更倾向于通过视觉信息和直接链接来获取信息。在具体行为上，点击行为的发生频率最高，占所有有效行为的55%。

[0073] 行为路径分析：通过对用户行为序列的挖掘，发现了一条核心的转化路径："T-01(标题) ->P-01 (产品外观图) ->P-02 (操作步骤图) ->L-01 (购买链接)”。约有32%的用户遵循了这条或类似的路径并最终点击了购买链接。用户的平均交互路径长度为4.8个元素。

[0074] 结果输出与应用：对创作者的优化建议：结果应用模块生成了一份分析报告。报告指出，参数表格TBL-01对用户的吸引力很低，建议将其内容可视化，例如改用一张信息图来呈现。同时，操作步骤图P-02和避坑指南正文P-05是高价值内容，建议将P-02的位置提前，并在P-05段落末尾也增加一个购买链接，以缩短转化路径。

[0075] 对推荐系统的赋能：相应地，结果应用模块将分析得出的用户兴趣标签（如对咖啡机、操作演示元素有高交互的用户）打包，通过接口提供给个性化推荐系统。当这些用户再次访问平台时，推荐系统可以优先向他们推送其他与咖啡机操作技巧或便携电器相关的图文或商品，从而提升推荐的精准度。

[0076] 通过与现有技术的对比，本发明实施例的有益效果显著。现有技术通常仅能提供素材页面的整体浏览量（1000次）和平均停留时长（62s）等宏观数据，无法告知创作者具体是哪个部分吸引或流失了用户。而本发明的方案，通过深入到元素级别的精细化追踪与分析，不仅将有效数据的占比从传统方法的约68.2%提升到98.8%（提升44.9%），分析维度也从3个宏观维度扩展到12个以上的细分维度（提升300%），最终指导内容优化后，该素材的链接点击转化率有望从18.3%提升至32.0%（提升74.9%）。此外，由于采用了Elasticsearch进行索引，复杂多维数据查询的响应速度可以控制在300毫秒以内，相较于传统数据库的秒级响应，效率提升了94%以上。

[0077] 本领域技术人员知道，除了以纯计算机可读程序代码方式实现本发明提供的系统及其各个装置、模块、单元以外，完全可以通过将方法步骤进行逻辑编程来使得本发明提供的系统及其各个装置、模块、单元以逻辑门、开关、专用集成电路、可编程逻辑控制器以及嵌入式微控制器等的形式来实现相同功能。所以，本发明提供的系统及其各项装置、模块、单元可以被认为是一种硬件部件，而对其内包括的用于实现各种功能的装置、模块、单元也可以视为硬件部件内的结构；也可以将用于实现各种功能的装置、模块、单元视为既可以是实现方法的软件模块又可以是硬件部件内的结构。

[0078] 以上对本发明的具体实施例进行了描述。需要理解的是，本发明并不局限于上述特定实施方式，本领域技术人员可以在权利要求的范围内做出各种变化或修改，这并不影响本发明的实质内容。在不冲突的情况下，本申请的实施例和实施例中的特征可以任意相互组合。< / script>

Claims

1. A method for tracking user behavior in graphic and textual materials, characterized in that, include: Step S1: The graphic and text materials to be analyzed are structurally split into multiple basic element units, each basic element unit is assigned a unique identifier, a lightweight behavior collection script is embedded in the graphic and text materials, and an interactive event listener is bound to each basic element unit. Step S2: Monitor and collect effective interaction behavior data between the user and each basic element unit in real time on the user terminal through the behavior collection script; Step S3: The collected valid interaction behavior data is preprocessed on the backend server to obtain the behavior sequence of a single user access session; Step S4: The behavior sequence is stored using a distributed database architecture, and a multi-dimensional index is built based on the key fields of the behavior data to support fast query and statistical analysis; Step S5: Based on stored data and multidimensional indexes, perform multidimensional analysis through a preset statistical analysis model, including calculating the interaction heat score of each basic element unit and mining user interests, preferences and behavioral paths; Step S6: Output the analysis results in a user behavior analysis report in a visual form, and simultaneously synchronize the structured user interest data to the downstream business system to achieve data-driven business optimization.

2. The user behavior tracking method for graphic and textual materials according to claim 1, characterized in that, The behavior collection script includes a core collection layer responsible for implementing the core logic of data collection, an event listening layer responsible for binding and listening to user page interaction events, and a data reporting layer responsible for packaging the collected data and sending it asynchronously to the backend server.

3. The user behavior tracking method for graphic and textual materials according to claim 1, characterized in that, In step S2, the effective interactive behavior must meet the preset judgment criteria, including at least one of browsing, clicking, staying, collecting, sharing, and downloading. Valid browsing requires that the element enters the visible area and the duration reaches the element type-specific threshold. Valid clicking requires that a valid interactive event is triggered, the click duration exceeds the threshold, and the coordinates fall within the element's valid interactive area. For images and long text elements, we also collect in-depth behavioral data, including zooming, dragging, and reading progress.

4. The user behavior tracking method for graphic and textual materials according to claim 1, characterized in that, In step S5, the interaction popularity score of the basic element unit is calculated by weighting the interaction frequency and interaction duration, and the calculation formula is as follows: in, and All are weights. Values ​​0.6 The value is 0.4, or can be dynamically adjusted according to business needs.

5. The method for tracking user behavior of text and image materials according to claim 1, characterized in that, In step S5, the multi-dimensional analysis also includes the overall interaction analysis of the materials, specifically the total number of material views, the number of effective users, the average page dwell time, the bounce rate, and the overall interaction coverage of elements. The user behavior path analysis uses a sequence pattern mining algorithm to reconstruct the typical interaction paths of users in graphic and text materials.

6. A user behavior tracking system for graphic and textual materials, characterized in that, The method for implementing any one of claims 1 to 5 includes: a structured processing module, a data acquisition module, a data processing module, a data storage and indexing module, a statistical analysis module, and a result application module; The structured processing module is used to split the graphic and text materials into basic element units, assign unique identifiers, embed behavior collection scripts and bind event listeners, and output structured graphic and text materials; The data acquisition module is a lightweight behavior acquisition script embedded in structured graphic materials, which collects effective interactive behavior data on the user terminal and reports it to the backend. The data processing module is used to preprocess the raw behavioral data to form a user access behavior sequence; The data storage and indexing module uses a distributed database composite architecture to store behavioral data and constructs a multi-dimensional index. The statistical analysis module performs multi-dimensional statistical analysis through a preset model, calculating element interaction popularity and mining user preferences and behavioral paths. The results application module generates a visual report from the analysis results and synchronizes user interest data to downstream business systems.

7. The user behavior tracking system for graphic and textual materials according to claim 6, characterized in that, The structured processing module integrates a parser and an identifier generator; The parser traverses the structure of the source document to identify basic element units; The identifier generator generates unique identifier information for each unit according to preset structured coding rules.

8. The user behavior tracking system for graphic and textual materials according to claim 6, characterized in that, The data processing module has built-in filters, converters, and a session manager; The filter removes invalid and redundant behavioral data, the converter transforms all behavioral data into a unified standard data structure, and the session manager aggregates the behavioral sequence of a single visit based on the user's anonymous identifier.

9. The user behavior tracking system for graphic and textual materials according to claim 6, characterized in that, The data storage and indexing module adopts a combined architecture of HBase distributed columnar database and Elasticsearch distributed search engine; HBase is used for high-concurrency writing and persistent storage of massive behavioral data; Elasticsearch is used to synchronize key fields of behavioral data and build multidimensional indexes, supporting complex queries and aggregation analysis at the millisecond level.

10. The user behavior tracking system for graphic and textual materials according to claim 6, characterized in that, The results application module includes a visualization submodule and an interface service submodule; The visualization submodule renders the analysis results into a web dashboard / analysis report that includes element heatmaps, indicator trend charts, and user path maps; The interface service submodule encapsulates structured user interest and preference data into standard interfaces for downstream business systems to call, forming a data-driven closed loop.

Images