Game Log Event Extraction Using NLP to Reduce Computer-Vision Load
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for extracting in-game events are inconsistent, computationally burdensome, and prone to errors due to varying game representations and updates, leading to unreliable output in external software applications.
Innovation Solution
Utilizing Natural Language Processing (NLP) to analyze text generated by game applications to identify in-game events, allowing extraction without requiring game code to conform to specific APIs and reducing the need for frame analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If computer vision techniques (OCR or object detection) are used to extract in-game events from gameplay frames, then in-game events can be identified, but significant computational burden is imposed that prevents real-time extraction
Solution Approach 1:
The patent extracts only the essential text information from gameplay frames using OCR, rather than performing full computer vision analysis. By isolating and processing only the text elements (event logs, notifications, scores) rather than analyzing entire frames with object detection, the computational burden is significantly reduced while maintaining event extraction accuracy.
Solution Approach 2:
The patent replaces the computationally intensive mechanical system of computer vision analysis with a more efficient text processing approach. Instead of using OCR combined with object detection algorithms that require heavy computational resources, the system uses OCR to extract text and then applies natural language processing rules to identify events, substituting the mechanical vision analysis with a lighter computational methodology.
2Reliability
If computer vision techniques are used to analyze gameplay frames, then in-game events can be detected, but the approach is sensitive to game updates and layout changes causing missed events or false positives
Solution Approach 1:
The patent introduces text as an intermediary layer between the gameplay visual information and the event detection system. By using OCR to convert visual game elements into text, and then applying NLP techniques to analyze this text, the system creates a language-based intermediary that is independent of visual layout changes. This text intermediary remains stable across game updates while still conveying event information.
Solution Approach 2:
The patent changes the parameter space from visual features (coordinates, colors, shapes) to linguistic features (text content, patterns, keywords). By transforming the detection problem from visual pattern recognition to text pattern matching, the system becomes less sensitive to graphical changes in game layouts and more adaptable to different game versions.
3Productivity
If APIs are incorporated into game code to communicate in-game events, then event data can be obtained, but game code must conform to inconsistent API representations across different games
Solution Approach 1:
The patent creates a universal text-based interface that copies the essential information function of APIs without requiring actual API integration. By extracting and processing text that naturally appears in gameplay (event logs, notifications), the system replicates the data acquisition capability of APIs while maintaining independence from any specific game's API structure or requirements.
Data Source
AI summary
In various examples, natural language processing may be performed on text generated by a game to extract one or more in-game events from the game. The system (e.g., a client device and/or server) may receive the text in the form of one or more strings generated by a game application. The system may then extract one or more in-game events from the text using natural language processing. The game may include the text in a message it sends to the system (e.g., using an Application Programming Interface (API)) and/or in a game log entry or notification. The text may be generated based at least on the game determining one or more conditions are satisfied in the gameplay (e.g., victory, points scored, milestones, eliminations, item acquisition, etc.). The text may be mapped to event templates, which may then be used to extract parameters of events therefrom.