Search project overview
Search project overview
Many people think that implementing a search solution is easy and can be achieved in a matter of days. While getting a basic solution to work can be done quickly, making that solution provide contextually relevant information to all end users requires more work.
This article outlines the steps you should follow to implement a successful, long lasting search solution relying on the Coveo Platform.
Know your end-user expectations
One common mistake in the early stages of a search project is to index all available content without first identifying the actual needs and expectations the end users may have. Before you even start working on your search project, you should be able to answer the following questions:
Question | Some possible answers |
---|---|
Who are your end users, and what are their search use cases? |
|
What information do your end users expect to find in their search results? |
|
What metadata do your end users expect to be able to filter their search requests or results on? |
|
How do your end users expect relevance to affect their search results? |
|
To gather such knowledge, a common practice is to carry out interviews and shadowing sessions with potential end users. Analyzing available usage data can also be helpful, especially if spending time with potential end users isn’t an option.
Locate and index content
Note
For more detailed information, see the Locate and index content section of this guide. |
End users often have a rather vague idea of where the information they need actually resides. In fact, they may only be aware that the content is accessible through some kind of portal. Early in your search project, you must identify the exact system (or systems) where the content your end users need is located (for example, SharePoint, Google Drive, Khoros Community, Salesforce, etc.).
The Coveo Platform offers a large selection of connectors which can be configured as sources to regularly pull items and permission models from original content repositories into an index. Typically, you should be able to use system-specific sources to index most, if not all of the content your end users require. If no system-specific source exists for a certain type of content repository you need to index data from, there are other possibilities (for example, using a generic source such as Database or Sitemap, or using the Push API).
Design the search experience
Note
For more detailed information, see the Design the search experience section of this guide. |
A complete search experience may include one or more search interfaces aimed at satisfying the end-user needs and expectations.
Among other things, designing your search experience may require you to:
-
Decide which filters and refiners should be available, and how the end user will interact with them (for example, through facets and tabs).
-
Create result templates to render certain types of items in a tailor-made fashion (for example, to highlight new items).
-
Develop custom components to meet specific end-user requirements (for example, to display a result count next to each tab).
While implementing a search interface entirely on your own is possible, doing so can be complex and time consuming. Thankfully, the Coveo Atomic library offers a set of customizable components which you can assemble to easily create feature-rich search interfaces.
Tune relevance
Note
For more detailed information, see the Tune relevance section of this guide. |
Indexing and querying content are merely the first steps of a successful search project. Tuning relevance is also an important and recurring task.
Manually tune relevance
The index provides built-in relevance through various ranking factors and weights.
You can also define your own relevance tuning rules by injecting query syntax expressions in search requests before they’re sent to the Coveo Search API, or by creating various query pipeline rules (these include query ranking expressions (QREs) and thesaurus rules, among others).
Examine usage analytics data
End-user interactions with a search interface can be recorded through the Coveo Usage Analytics (Coveo UA) service.
You can leverage UA data in reports to learn what your end users are searching for, what they’re finding, and also what they’re not finding (which can help you identify content gaps). You can also derive end-user behavior from this data, which lets you make informed search experience design choices and relevance tuning decisions. Finally, the Coveo Machine Learning (Coveo ML) service relies on usage analytics data to improve the output of its models.
Activate machine learning
The Coveo Platform is a relevance engine that leverages search to understand and learn from end-user intent so that it can provide increasingly relevant interactions. This implies that when enough reliable usage analytics data is available, the Coveo ML service can dramatically improve the relevance output of your search project.
-
Based on the current query, an Automatic Relevance Tuning (ART) model can increase the ranking scores of items that are most frequently opened by end users performing similar search requests.
-
A Query Suggestion (QS) model can provide basic query expression (
q
) completion suggestions which returned relevant results in the past. -
A Content Recommendation (CR) model can yield a list of items that were viewed by other end users in a similar context.
Continuously improve the solution
A search project is never done. You should continuously examine usage analytics data to fill content gaps, improve the search experience design, and fine tune relevance. The Coveo Administration Console offers tools to do this in a most efficient way.
What’s next?
The Evaluate the project time frame article presents the key activities of the IDEAS methodology and outlines the factors to consider when planning a project.