Machine Learning Tag Classifier for E-Commerce Product Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing diversification of commercial products in electronic commerce poses a challenge in efficiently classifying and attaching appropriate tags to products based on text and image information, as existing technologies struggle to effectively determine relevant tags from a vast array of options.
Innovation Solution
An information processing apparatus and method that utilizes a processor to reference a correspondence list between product genres and tags, training a machine learning model with a classifier to determine output values for relevant tags based on embedded expressions in product information, excluding output values for mismatching tags, thereby optimizing the tagging process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning model trains on all tags, then comprehensive tag coverage is achieved, but training efficiency and accuracy deteriorate due to inclusion of mismatching tags
Solution Approach 1:
The system performs preliminary classification by product genre before tag determination. A correspondence list storing genre-tag relationships is prepared in advance, and the processor determines which tags are relevant based on the product's genre before training the machine learning model. This preliminary action filters out mismatching tags, reducing training time while maintaining accuracy.
Solution Approach 2:
The tagging process is segmented into two stages: first, genre-based filtering to identify relevant tags using the correspondence list; second, machine learning-based determination only for those filtered tags. This segmentation separates the tag selection process from the tag determination process, improving efficiency by excluding mismatching tags from training.
2Adaptability or versatility
If machine learning model considers all tags, then all possible tags are evaluated, but processing complexity increases due to mismatching tags
Solution Approach 1:
The correspondence list is prepared in advance to establish which tags are applicable to which product genres. Before the machine learning model processes tags, the processor uses this pre-prepared list to filter and determine only the relevant tags for the given product genre, reducing processing complexity while maintaining versatility.
3Measurement precision
If loss function includes all output values, then complete model evaluation is performed, but training efficiency decreases due to mismatching tags
Solution Approach 1:
The processor determines relevant tags based on product genre and the correspondence list before training. The loss function is then constructed to include only the output values corresponding to these relevant tags, excluding mismatching tags. This preliminary filtering maintains measurement precision while improving training efficiency.
Solution Approach 2:
The system extracts and excludes output values related to mismatching tags from the loss function calculation. By taking out only the necessary output values corresponding to relevant tags, the training process becomes more efficient while the model evaluation remains precise for the applicable tags.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Disclosed herein is an information processing apparatus including a processor and a memory. The processor references a list indicative of correspondence between a genre of products in electronic commerce and tags to be attached to the products, thereby determining at least one tag not corresponding to the genre. The processor performs a process of training a machine learning model including at least a classifier that determines an output value of each of the at least one tag related to the products on the basis of an embedded expression in product information on the products, the training process being performed on the basis of a loss function of the at least one output value excluding the output value related to each of the at least one tag not corresponding to the genre of the products.