Search Tuner

You click Enable boost search for articles to make certain articles show higher in the search results based on the article attribute you want to boost. A boost score of five means the attribute must match the search string. Any score that is lower than five is a boost but not a must-match.

You can also change the matching logic between AND, OR, and PHRASE:

• AND means all words in the search string must match.
• OR means not all words in the search string must match.
• PHRASE means all words must match in the order searched.

The search method let you select one of the following options:

• CONTAINS, which matches parts of the word.
• EXACT, which matches the entire words.
• FUZZY, which matches the start of the word but allows a mix of characters to compensate for misspelled terms.

The Search configuration landing page is also where you enable NLP skip words. This feature requires you to specify all skip words in a Prediction Model (see kmsearch_skipwords in the Predictive Models section).

Categories are a way of grouping content into large buckets of similar content. The exact structure of your categories (in the Taxonomy) is a business decision depending on how you want your application to use data. As a best practice, begin with the highest level of categories you can, and then break it down if needed. Each article belongs to one or more categories.

Tags are the next level down from categories. They are keywords or terms representative of the content within the article itself, and each generally has between one and five tags. The tag structure is also a business decision that you adjust depending on your needs.

The Search Tuner analyzes the search string and extracts keywords from it. Based on the Predictive Models you build, the keywords link to certain taxonomy categories or article tags. The search then returns any article matching either the categories or tags.

The Search Tuner shows extracted categories inside green squares and article tags inside grey squares. Additionally, the content of the article from the top search result is displayed on the right side of the screen. The categories and tags of this article are displayed in a similar way. When you click on a different article, the display shows the content of that article instead.

The following example shows the extracted categories and tags under the search field, as well as the selected article on the right side of the screen:

The Search Tuner uses five Predictive Models to function properly; four of the models connect to search, and one connects to articles:

1. kmsearch_categories: Contains all keywords that link to a certain taxonomy category during a search.
2. kmsearch_tags: Contains all keywords that link to a certain article tag during a search.
3. kmsearch_skipwords: Contains all words that the system ignores during a search (for example, above, about, the, after, and others).
4. kmsearch_replacements: Contains a two-level structure that instructs the search tuner to replace certain words with other, predetermined words (for example, colour with color or favourite with favorite).
5. kmarticle_tags: Contains keywords that cause a certain tag to be assigned to an article if auto-tagging is enabled.

Both kmarticle_tags and kmsearch_tags must populate correctly. If no article has a specific tag, matching a keyword from the search string to that particular tag does not return any result. You must update article tags after creating a new tag.

If a search does not return results, you can change either categories or tags according to your business needs.