If you, as the Klevu user, have multiple representative types of products in your catalog; e.g. mobile phones and mobile phone accessories, you are likely facing an issue where searching for the term “iPhone 7” is returning accessories at the top, as opposed to the mobile phones which represent the optimal results.
This is a very common issue - below we define the issue itself and also the solution.
If we continue with the iPhone example, you likely have the two products from the above categories:
iPhone 7 32GB, White (from the mobile phones category)
iPhone 7 Case (from the mobiles accessories category)
Here, when you search for the term “iPhone 7” (more detail on native rankings can be found here: How does Klevu rank products?), our algorithm (without any training), will look for the products with the words “iPhone” and “7” in the indexed product fields (e.g. name, description, category, attributes). At this stage, without any training in place, the products where these keywords appear in product names are likely to appear at the top. The problem in the above case is that the term “iPhone 7” appears in both the products’ names and it isn’t very clear which product should appear at the top.
In our higher plans, we use natural language parsers to identify the intent of a shopper and our algorithm has the capability to understand that that the user is looking for the device. However, in the absence of such intelligence, the system requires external support to solve the issue.
The resolution
If you’re on a lower package of Klevu, the most obvious fix is to apply a boosting rule to increase/reduce exposure for the selected items on more generic queries. You can use the bulk product promotions feature to deboost products featured in the mobiles accessories category. For example, you can assign a 0.75 boosting score, which would ensure that both sets of items are included, but products from the mobile accessories will be multiplied with the 0.75 boost score, resulting in lower relevancy scores for the mobile accessories.
Another option would be to assign the mobile phone devices as hero SKUs against an individual or multiple keywords, using the “keyword-based product promotions” feature.