Introduction to Smart Snippets

 

With a typical site search experience, users often face a set of results that may contain the answer they're looking for without knowing exactly where the information they want is located. Pinpointing which document in the search results contains the right information, then locating that information within a document can be frustrating, plus it adds time to the search process.

Jason Mlyniec, Director of Customer Onboarding at Coveo, notes that this common search experience often leaves users wondering how to phrase their queries to get results that directly answer their questions. Smart Snippets, a Coveo machine learning model, aims to solve these problems.

 

How Coveo's Smart Snippet Model Works

 

Mlyniec provides a comprehensive overview of Coveo's Smart Snippet model, which leverages advanced deep learning technology to identify and extract the most relevant information from your indexed content.

 

The model identifies headings within the content and embeds them, along with the user's query, into a vector space. It then calculates a similarity score between the query and the headings to determine whether a snippet should be displayed.

 

This process occurs after the initial search API is applied, refining the top 10 results. The model essentially seeks to match the user's query with the most relevant headings in a document. The caveat is that content must be properly structured with appropriate HTML tags (H1, H2, H3, etc.).

 

Some key points about the Smart Snippet model:

• Unlike other Coveo ML models, it doesn't rely on usage analytics to train itself — it learns from the content directly
• Semi-structured or structured HTML content works best
• It requires a Coveo Enterprise license

 

The key here is relevance. Smart Snippets identify the most relevant snippets, then presents them prominently in the search results. This solves the problem of users having to scroll through results then guess which one to click on. The model also enables users to quickly find information without having to dig through entire documents.

 

Smart Snippet Implementation

 

Implementing the Smart Snippet model is a straightforward process that can be completed in just a few steps. Mlyniec outlines the process, beginning by defining the document source within the Coveo organization you’ll be using to generate smart snippets. Next, you'll build the Smart Snippets model in the Coveo Admin Console and associate it with a query pipeline.

 

The last step is adding a couple of lines of code to your search interface to enable the smart snippets and the "People Also Ask" component. For a detailed walkthrough of the implementation process, be sure to watch the full webinar.

 

Advantages of Smart Snippets

 

Smart Snippets pack a powerful punch when it comes to improving search experiences and driving user satisfaction. Mlyniec highlights some compelling use cases for this feature, including support portals, chatbots, and search interfaces. By pulling relevant pieces of content directly into the search results, Smart Snippets help users avoid the hassle of digging through long documents.

 

Smart Snippets are incredibly effective customer service tools since they give users fast, accurate answers to questions with reduced effort. They also improve case deflection since they help customers solve their own issues, reducing the burden to service teams.

 

Smart Snippets go beyond efficiency. They also create a more engaging, intuitive search experience which includes pulling in related "People Also Ask" suggestions. This encourages exploration and discovery, turning a basic search into an opportunity to learn something new.

 

Introduction to Case Assist: Case Classification Model

 

The webinar shifts gears to explore Coveo's Case Assist Model, presented by Lipika Brahma, Customer Success Architect at Coveo. Case Assist is designed to reduce case creation volume and ensure that cases are routed to the appropriate teams.

By leveraging natural language processing (NLP) models to analyze the case subject and description, Case Assist predicts the relevant product, category, and topic, saving customers time and effort when submitting cases.

 

Coveo Case Deflection

 

Case Assist helps improve case deflection by reducing the need for agent intervention. The case deflection component uses two Coveo ML models including Intelligent Term Detection (ITD) and Automatic Relevance Tuning (ART) to provide contextual recommendations.

These models analyze user behavior and case content, using this information to identify and highlight the most relevant solutions. This is true even for complex issues presented in lengthy case descriptions.

 

Case Assist Implementation

 

Brahma walks attendees through a live demo demonstrating how to implement Case Assist, a process that includes:

• Defining the case source
• Selecting the fields for analysis (like subject and description)
• Specifying the fields for prediction (like product and category)

 

Viewers get a step-by-step walkthrough of how to configure Case Assist in the Coveo Cloud platform, including how to create and train the Case Classification model. A prerequisite for Case Assist is having sufficient high-quality case data for the model to learn from. Specifically, you need a minimum of 500 closed cases for optimal results.

 

Case Assist is an API-based solution, typically used along with the case creation UI. Two ML models — Case Classification and Document Recommendations — work in tandem to enable this feature. Brahma focuses on the former.

 

Advantages for Case Classification

 

Case Assist and its Case Classification model help reduce friction and improve search. Case Assist simplifies the case submission process for customers, for example, by predicting case fields like product, category, and topic.

Case Classification can significantly boost support team efficiency by ensuring cases are routed to the right agent or team. Case Classification uses ML to continuously improve its predictions, learning from each interaction. The more cases it processes, the smarter it gets, driving even better results for customers and support teams.

 

How Case Classification works Within Case Assist

 

Brahma explains the two types of Case Classification that power Case Assist: case similarity and context recognition.

 

Case similarity looks at existing cases and suggests categories based on similar ones, ensuring that new cases are properly categorized. Context recognition uses an NLP model to understand the context and nuances of the case content.

 

The NLP model excels at analyzing longer cases, identifying key information even in complex descriptions. This model works well for organizations with support cases that contain more than eight words – which is most businesses. By leveraging case similarity and context recognition, Case Assist ensures that cases are accurately classified and routed to the appropriate teams.

 

To learn more about setting up and optimizing Coveo's Smart Snippets and Case Assist models, be sure to watch the full webinar. You'll come away with a clear understanding of how to leverage these powerful machine learning tools to enhance your search experience, streamline customer support, and drive better business outcomes.