Improving Search Results

If your company research brings up too much noise or too many false positives, try these steps to improve your results. For this article, we’ll work through the process of improving the search results for The Ordinary, a beauty brand.

If you'd rather watch a video walk-through, check it out here.

1. Start with the Knowledge Graph

Whenever possible, start with an entity in the SESAMm knowledge graph. Entities in the knowledge graph are not disambiguated, but they are already mapped to keywords, giving you a head start.

2. Add Additional Keywords

From there, add additional keywords of interest to you. For example:

All versions of the entity’s name: The Ordinary Skincare
Product keywords or names: The Ordinary shampoo, The Ordinary cleanser, The Ordinary multi-peptide, The Ordinary natural moisturizing factor, The Ordinary serum,
Keywords in other languages: The Ordinary huile, The Ordinary crème hydratante

If you need some inspiration for keywords, try the website of the entity you are researching, it’s usually a great source.

3. Evaluate the Search Results - and Identify Ambiguous Keywords

After adding additional keywords, run the search and evaluate the results. If you have too many false positives, start looking for ambiguous keywords and keywords to exclude. A good place to start is with the Trending Topics tab.

If you find an article that doesn’t fit, take a look for keywords that you could exclude to improve the results. For example, below we see “out-of-the-ordinary” as a possible keyword to exclude.

Another way to identify problematic keywords is to look at the trends. If there’s a spike in article volume, dig into it and see what is driving the spike.

Once you have identified the problematic keywords, move on to disambiguating and excluding them.

4. Disambiguate

Our example shows that the keyword “The Ordinary” is ambiguous. Using disambiguation, we can specify additional keywords that need to be mentioned in the same article. For example:

Market or product-related keywords: skincare, cleanser, multi-peptide, beauty, serum, pores, primer, exfoliator, face oil, toner
Brand-related keywords: DECIEM (which owns The Ordinary)
Split brand names: This doesn’t apply to this search, but you could also try to split brand names. For example, if you were researching Picard Surgelés (a frozen food store in France), you might use Picard as a keyword, but disambiguate it with “surgelés”.

💡What is Disambiguation?

Disambiguation is an important tool for improving your search results in Open Search. If your results turn up too many false positives, one of the keywords may be ambiguous - it could have several possible meanings. For example, Apple could refer to both the fruit or the company, while Patagonia could refer to the region or the company. To distinguish between them, we use disambiguation.

Disambiguation involves adding keywords that must be co-mentioned with the ambiguous keyword when a document is screened. Using our previous example, disambiguating the keyword “Apple” might mean adding “computer” and “iphone” as co-mentions. Another way to think of disambiguation is in terms of a Boolean search, written out it would look like this: “Apple” AND (“computer” OR “iphone”).

Lastly, it’s important to note that disambiguation keywords are searched for in the text field only, not in the title.

5. Exclude Keywords

The final step is to exclude keywords that are consistently mentioned in false positives. In this case, we’re going to exclude, “out-of-the-ordinary” and “resorts”.

If you have a lot of keywords to exclude, it may be more efficient to reduce the number of keywords added in step 2. For very ambiguous or common keywords, it is often simpler to start with a more restricted search and slowly expand it.

6. Save your Work

Once you’re happy with your search results, don’t forget to save the entity. You’ll be able to easily access it in the future and share it with your colleagues.