Because eCommerce title space is limited, we need to know how to prioritize which words to recommend including. The approach I used was to analyze the list price and search engine performance report to score every word used in product titles and measure its positive, neutral or negative impact on price and conversion performance. We then used this data combined with the structured attributes and product descriptions to suggest building product titles that would convey the most value.
I used both a custom-coded corpus aggregration and a machine-learning random forest models.