Abstract
Traditional user interaction logging processes typically record discrete events based on user interactions with a user interface of an application. The user interactions may include user clicks or selections of interactive user interface elements (e.g., click events) and/or user impressions of content (e.g., impression events) provided by the user interface. Interactive user interface elements may include, but are not limited to, buttons, selectable icons, advertisement cards, and content cards. For example, a user impression may occur when a content item (e.g., an advertisement, an image, a video, etc.) is rendered in the user interface (e.g., in a respective content card) regardless of whether the user interacts with the content item (e.g., selects or clicks on it). The user impression and/or the user selections may be considered fundamental metrics of a measure of the reach or frequency of content delivery to the user.
In some implementations, the recorded discrete events may provide a somewhat limited view of the complete user engagement with the user interface of the application. The recording of the discrete events may result in basic metrics that often fail to capture a nuanced user decision-making process and/or an implicit user intent that may occur between actions. For example, an implicit user intent that may occur between actions may be a user hovering over elements and/or content (e.g., interactive user interface elements, content items, etc.) in the user interface and/or a focused browsing of the user interface by the user.
The disclosed technology may classify and/or group sequences of user interactions with the user interface of the application into one or more high-value behavioral categories. For example, impression events and click events may be stitched together to identify patterns of usage and/or interaction of the user with the user interface. The identified patterns may include, but are not limited to, a fixation of the user on specific items or content in the user interface, row-level browsing of the user interface, and interest area transitions within the user interface. For example, a delay between an impression event and a click event may be classified as a hover event.
In some implementations, test user accounts may include a basic understanding of how a user may interact with a user interface. The test user accounts may use the basic understanding to trigger a pop-up menu on the user interface requesting feedback from the user related to the user interactions with the user interface. Initial heuristic sequential machine learning (ML) models for use in identifying and classifying user engagement with the user interface may be refined, updated, or fine-tuned using the feedback gathered by test accounts. In addition, or in the alternative, subsequent large-scale user surveys may be used to further train the sequential ML models. The incorporation of the identified and classified user behaviors with weighted values in the refining, updating, fine-tuning, and training of the sequential ML models may result in recommendation models for a user of an application where the user affinity towards preferred content is more accurately represented. The disclosed technology may improve the predictive quality and responsiveness of recommendations by leveraging subtle interaction signals of the user with content in the user interface.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Chatterjee, Tamojit and Mishra, Kanishk, "Advanced User Interaction Logging for Recommendation Model Enhancements", Technical Disclosure Commons, (February 10, 2026)
https://www.tdcommons.org/dpubs_series/9307