One issue the customer faced was inconsistent or inaccurate product listing entered by multiple vendors or third-party sellers as a result of a lack of knowledge, duplication or oversight. The other problem was manually validating metadata tags which was prone to human errors like spelling mistakes, incorrect mapping, duplication of information and oversight. Data had to be exported to and imported from various sources, and from a wide range of sellers who often have different formats, vocabularies, and methods of documenting product information. Generic product content was also impacting sales and engagement.
Inconsistent product listing
With Vue.ai’s state-of-the-art image classification capabilities, deeper attribute-related information that describe products were also extracted. The Natural Language Processing (NLP) capabilities classifies and extract features from differing sources of metadata about the product and its attributes.
Discrepancies in information from various sellers:
Automated product tagging audits data entries to identify missing data, incomplete metadata or duplication of images across sellers. It helps reconcile visually and textually extracted attributes. This highlights inconsistencies and/or boost the confidence and accuracy of attributes that have contextual cues from the different sources.
Manual Quality Assessment (QA):
With QA, algorithms improved, improving accuracy over iterations of feedback. The algorithms also identified confidence cut-off beyond which margin of error was acceptable, reducing the need to QA
With the current tagging process & efficiency, Vue.ai was able to correct 51.82% of total tags processed in the live site—i.e., their existing product catalog, improving data quality in the process.
Case-specific product tagging:
In addition to the rich attributes extracted, attributes are mapped to specific taxonomy along with confidence values for the extracted attributes, improving the predictions made.
For this client, Vue.ai's automated product tagging solution predicted tags for over 3 Million products across categories over a period of 3 months. Over this period, it added 69 new attributes, values covering 18 attributes & 21 Product types across 3 categories.
