Getting Started with Site Search
Welcome to SearchStax Site Search.
SearchStax Site Search is a game changer that makes it quicker, easier and less expensive for companies to deliver relevant and personalized site search experiences on their websites. Site Search gives control over the search experience to your business and marketing teams – and reduces IT bottlenecks.
Once Site Search is integrated into your website, your business teams take over and use the point-and-click interface to:
- Configure Site Search with Results, Facets, Stopwords, Synonyms, Spell Check and Auto-suggest
- Gain Insights with Search Analytics using the Dashboard, Searches, Items and Power Search
- Optimize search results relevance with Ranking, Rules and Promotions
- Personalize results by a different optimization model depending on the audience value
- Gather feedback with the Search Feedback widget
Explore the Site Search Product Documentation using the navigation in the left panel.
This page explores the general concepts behind Site Search. For instance, there are two user personas served by this document, reflecting their two very different levels of interaction with Solr and Site Search.
- Developer – This use is responsible for Setup, Data Modeling, Data Loading and Integration with the Client Application.
- Business User – This user views Analytics data, configures Search Features, creates Search Relevancy models, and tests the results.
The goal of Site Search is to empower a Business User to shape and control the website search experience. The Business User interacts exclusively with the Site Search user interface, making point-and-click modifications in the presentation of search results.
Before reaching that point, however, the Developer must create a Solr deployment that can respond to user queries.
The initial steps involve the Developer. There must be a Solr search engine in the background, with a customized search page that sends user events and feedback to a Site Search App. The details of these tasks are shared in the Developer Getting Started page.
Once the infrastructure is in place, and the Site Search App has accumulated several weeks of data, control transitions to the Business User. The project usually pursues the following milestones:
- Examine the recorded data in Site Search (keywords, results, and click-through events). Identify opportunities to improve the search experience.
- Use Site Search to adjust:
- Use Site Search to create Relevance Models.
- Evaluate each Relevance Model by comparing search results.
- Possibly promote a Relevance Model to be the default system behavior.
Site Search Application
The first step in training your search engine is to take a detailed look at its current behavior. For this the Developer creates a Site Search Application (App) and connects it to a Search Page that reports user actions and feedback.
The Site Search App lets you record the history of searches, clicks, revenue, searches per session and user feedback over a period of time appropriate to your business. At a lower level, the App summarizes popular queries and critical metrics such as click-through rate, average click position, mean reciprocal rank, and frequency of no-results searches. See Analytics – Searches.
The Site Search App shows you how actual client searches translate into click-through events. It becomes a repository of queries and click events to drive experiments while you tune the search behavior. See Search Experience – Search Configuration.
Search Experience Manager
Once you know what your search engine is doing, you’ll want it to do something differently. Site Search gives a Business User direct control over the following parts of the search experience.
Results and Display
Index records have multiple fields, not all of which make sense to the public. The Site Search lets you choose which fields to show in the search results. You can order the fields to put the important ones at the top. It’s a point-and-click operation.
You can also control the number of results to present and optionally turn on hit-highlighting, a feature that emphasizes words that match the user’s query. See Results and Display Tab.
Site Search lets you pull up a specific search result and view the query keywords that led to click-through events. Sometimes quite different keywords can all lead to the same outcome. For example, people who purchase a “tank top” might have searched initially for “shirt,” “blouse,” “top,” or “singlet.”
Site Search lets you define some of the keywords to be synonyms. When a user enters a keyword, the search engine automatically includes synonyms in the search. This widens the search to include all records that use any of the terms.
Synonyms are defined in pairs, and can be one-way (A implies B) or reflexive (A and B imply one another). See Synonyms Tab.
Some fields contain clear, discrete values that make natural selection lists, such as a field that contains the names of the three primary colors. These lists are called “facets,” and are used to filter the search results.
The best facets are those that have just a few values (5 to 7, typically), so the user can see them all at once.
Site Search lets the Business User pick fields to use as facets, determine how many choices will be offered in each list, and define the order of the facet lists on the search page. See Faceting Tab.
It is a waste of memory and CPU time to search for words like “and” and “the” because they appear in virtually every record in the index. Most search engines filter out these “stopwords” before the search.
Site Search makes the stopwords visible to the business user and permits adding and removing words from the list. See Stopwords.
When users misspell or mistype a query keyword, they can miss out on your services and products. Spell checking compares the user’s input with a dedicated dictionary that you provide or with terms in the index of a specific collection. Either way, when a close match is encountered, Solr searches for both the original and the corrected terms. See Spell Check Tab.
To help users enter their search keywords, Site Search can maintain a collection (a Solr index) of terms that have been entered before. The Auto-Suggest feature returns a list of keywords that match the lettter-by-letter input while the user types in the search field. The user can click on one of these words to complete the search. See Auto-Suggest Tab.
Search Relevance Models
Site Search lets us create named sets of search configurations known as “relevance models.”
A relevance model lets us adjust which content fields will be included in the match score, and how much a match in each field will contribute to the final total.
Site Search lets the Business User select which fields to use when searching. (This is not the same as selecting the fields to display in the search results.) By narrowing the target fields, the user can speed up the search and make it more focused on critical content.
See Search Fields Tab.
Site Search lets you set the relative importance of the search fields, so that a keyword match in one field counts more heavily than the same match in a less-important field. A match to an author’s name might get more weight than one to a document title, which in turn might have more weight than a match in a document description.
Site Search lets you set the importance of your fields using slider controls. There is no programming involved. See Ranking Tab.
Rules are if-then propositions that let us respond to search keywords by boosting specific match fields or filtering the search results to items that include a specific field-value. For instance, if the incoming query includes the word “security,” then multiply the content field score by 10. See Rules Tab.
When the user searches for a specific keyword, sometimes the Business User wants to push one of the search results to the top of the list of matches. This is a “promotion.”
The Site Search lets us promote one or more items to the top of the list, whether it would normally match the triggering keyword or not. See Promotions Tab.
A typical Site Search App records thousands of real-life user queries, each bundled with all of the facet selections and click-through events that were part of the search.
Site Search lets us experiment by running searches against multiple relevance models. We can compare the original results and the new ones, summarized as changes in:
- Average Click Position: This is the mean click-through position in the search results.
- MRR: Mean reciprocal rank, a weighted mean of click-through position. Ranges from 0 to 1.
- No Result Search %: Percent of queries that returned zero results.
If the new results improve on the old ones, we can adopt those changes and start again.