Action Clustering for Social News Feed Relevance

The patent segments actions into distinct action groups based on their relationships and attributes. Each action is evaluated independently against a template action, and related actions are clustered into separate groups. This segmentation allows the system to manage complexity by handling actions in discrete, organized units rather than as a monolithic collection.

The patent changes the parameter of action organization from chronological or random ordering to attribute-based clustering. By computing scores based on attribute matches between actions and template actions, the system transforms the organizational parameter from simple time-based sequencing to a multi-dimensional attribute comparison framework, improving relevance while managing complexity through systematic parameter evaluation.

The patent applies partial action by selecting only the top-ranking actions that exceed a threshold score when constructing news feeds. Rather than processing and ranking all possible actions equally, the system computes scores for candidate actions, ranks them, and selects only those that meet a predetermined threshold or top-N criteria. This partial processing approach maintains high ranking accuracy for selected actions while reducing overall computational time and resources.

The patent changes the computational approach from exhaustive comparison to threshold-based filtering. By establishing score thresholds and selecting actions that meet or exceed these thresholds, the system transforms the problem from processing all actions with equal depth to focusing computational resources on actions that demonstrate sufficient attribute匹配, thereby improving accuracy for selected actions while reducing total processing time.

The patent merges multiple related actions into unified action groups that are then presented together in news feeds. By combining actions that share common attributes or relationships (such as actions related to the same event, person, or topic), the system creates coherent content clusters that enhance user engagement. The merging process is managed through attribute-based clustering, which organizes diverse content into meaningful groups without requiring manual curation.

The patent changes the content organization parameter from diverse, unstructured presentation to attribute-based clustering. By evaluating actions against template actions and grouping them based on attribute matches, the system transforms the organization task from a complex manual sorting problem to a systematic parameter-based classification process. This allows diverse content to be automatically organized into coherent groups that are easy to present and consume.