Suspicion Classifier for E-Commerce Search Relevance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
E-commerce sites face challenges in providing relevant search results due to improper promotion of item listings, including duplicate listings, misleading data, and shill activities that artificially increase an item's visibility and attractiveness.
Innovation Solution
A classifier system is developed to identify suspicious item listings by extracting relevant data, training a model to determine suspicion scores, and adjusting the visibility of listings based on these scores, preventing suspicious listings from being posted or prominently displayed in search results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Illumination intensity
If item listings include promotional data or features to increase visibility, then item prominence in search results is improved, but search result relevance deteriorates due to improper promotion
Solution Approach 1:
The system performs preliminary classification of item listings as suspicious or not suspicious before they are displayed in search results. By pre-identifying suspicious listings using trained classification models, the system prevents improper promotional content from affecting search result relevance, while still allowing legitimate promotional items to maintain their visibility.
Solution Approach 2:
The classification model acts as an intermediary between item listings and search result display. It evaluates promotional features and data to determine whether listings should be suppressed, thereby mediating between the desire for item visibility and the need for search result relevance by filtering out only suspicious promotional content.
2Reliability
If suspicious item listings are suppressed to improve search result quality, then search result relevance is improved, but the complexity of the system increases due to classification requirements
Solution Approach 1:
The classification system is trained using automatically generated training data from item listings, seller information, and buyer information without requiring manual annotation of each listing. The system serves itself by using its own operational data to train and improve its classification models, reducing the complexity associated with external data preparation and expert labeling.
Solution Approach 2:
The system uses multiple classification models that can be trained with different parameters and features (item data, seller data, buyer data, interaction data). By adjusting and changing these parameters based on training results, the system can optimize its complexity-relevance tradeoff without requiring a complete system redesign.
3Measurement precision
If classification models are trained with extensive data to improve detection accuracy, then suspicion detection precision is improved, but the time and resources required for training increase
Solution Approach 1:
The system performs preliminary training of classification models using accumulated historical data before deployment. By pre-training models with extensive datasets in advance, the system achieves high detection precision without requiring extensive training time during operational phases, as the models are already trained and ready for rapid suspicion detection.
Solution Approach 2:
The classification models can be continuously trained and retrained using new data that accumulates over time. This continuous training process allows the system to maintain and improve detection precision progressively without requiring complete retraining from scratch, thereby reducing the time loss associated with periodic model updates.
Data Source
AI summary
A system and method of identifying suspicious item-related features are disclosed. In some embodiments, a new item listing is received. The item listing may correspond to a request to list an item for sale on an e-commerce website. Item-related data for the item listing may be extracted. The item-related data can be used by a model to classify the new item listing. The model may be trained on data comprising identifications of which item listings in the sample of item listings are suspicious and identifications of which item listings in the sample of item listings are not suspicious may be received.