Relevance

Core Settings

Relevance optimizes how well the search results match the user’s intent and query. Increasing relevance means happier users through better findability of products and documents.

Relevant search results are important for:

• E-commerce sites, where relevant results drive conversions and revenue.
• Support sections of sites, such as help pages, where relevance reduces customer frustration and complaints.
• Enterprise search, where relevance boosts productivity and knowledge discovery by helping employees find what they need to perform their work or solve a problem.

Query profiles

When your site has multiple query profiles, you can select which profile you want to configure using the query profile selector at the top of the Relevance page:

Query profiles enable you to manage multiple search experiences within a single site. Each query profile represents a distinct search configuration, including a query pipeline and collection.

The query profile selector only appears when your site has multiple query profiles configured. If you have a single query profile, the selector is hidden and all settings apply to that profile automatically.

You can create and manage query profiles in Lucidworks Search or self-hosted Fusion, or by contacting Lucidworks.

If two query profiles point to the same query pipeline, changes made to one query profile’s Core Settings will impact the other.

Neural Hybrid Search

Lucidworks Core Settings includes the option of using Neural Hybrid Search (NHS) to take advantage of advanced searching capabilities that combine lexical search and semantic search, enabling exact and partial keyword matching, phrase detection, and search results that understand the natural language users might enter in the search box. To improve the relevance of results, you can also set values to assess how closely results match and number of results returned.

Weight balance between lexical and semantic search

Use the Weight slider to decide the relative importance of lexical and semantic search for your search results. As you increase the weighting of one type of search, the weight of the other type of search decreases.

Depending on the type of content you have, you can adjust the slider to accommodate the information you want the user to see.

For example, a legal, medical, or financial website might have industry-specific words that benefit from a more lexical approach, while an e-commerce website might want to show users similar products and will benefit from giving a greater weighting to semantic search.

NHS uses both of these to optimize the search experience.

Fine tune the results

To improve results, evaluate the specific needs of your use case to determine the appropriate balance between lexical and semantic search.

Begin testing by starting in the middle with 50% lexical and 50% semantic and perform a series of queries. If the results do not match your expectations, then adjust the slider.

B2C

In a B2C context such as a grocery store chain, if a user searches for “ice cream” and the results prioritize “bag of ice” and “sour cream,” the semantic weight should be increased. This helps return contextually relevant items like “waffle cones” and “chocolate syrup” that align with the shopper’s intent.

Financial services

In contrast, on a financial services site where a user is researching “bonds,” overly broad results like CDs, mutual funds, or ETFs may dilute relevance. Increasing the lexical weight ensures the top results remain focused on bonds.

Adjust the weighting strategy to match user intent and the goals of your business.

Adjust the weights for your specific use case, and validate that the search results become more relevant for the users.

Control how closely records match and the number of records returned

Two settings determine the number of records returned:

• Vector Similarity Threshold: This threshold sets the minimum of how closely a document must match your query’s meaning to be included in results. Enter a decimal value between 0 and 1.

  • A higher minimum value returns only results that strongly match the query’s meaning. Assessing possible matches to return strong results may increase latency (processing response time).

  • A lower minimum value returns more results because results do not have to match the query’s meaning as closely.

  • Results that are not semantically close enough to the query’s meaning are filtered out before results are re-ranked.

• Vector Result Count: This value specifies how many candidate results the vector (semantic) search retrieves before the results are merged and re-ranked with the lexical results. Enter an integer between 0 and 2000.

  • A higher value assesses more semantic candidate results, which improves relevant results, but may increase latency (processing response time).

  • A lower value assesses fewer semantic candidate results, which may decrease relevant results, but also takes less processing time.

These fields map to the Min Return Vector Similarity and Number of Vector Results fields in the Neural Hybrid Query stage settings.

Sort

Sort is how search results are ordered and can be set according to relevance, which is the degree to which each item matches a user’s query based on search settings.

You can change the default sorting order that is used for results by arranging the tiles in the Sort section. This lets you sort first by relevance, then by date for items with identical relevance, or sort first by date, then by relevance for items with identical dates.