Knowledge Base Search – How does it work?

This is a question that several of our customers have asked us, when they start to build their own knowledge base of answers, to enable customer web self-service.

The knowledge base search (knowledge foundation) is not a simple mechanism. That is why it is so powerful, intelligent, dynamic and self-learning.

A picture is worth a thousand words, so I decided to put together a diagram that depicts the process, and below leave you with a few definitions to be better understand the different components.

oracle_service_cloud_kb_search.png

Weight

When a search is performed, each keyword and/or phrase entered by the customer is compared to the contents of the answers.

The Weight is a numerically calculated value, based on the number of occurrences, capitalisation, and location of a word. It is equal to the sum of the weights of all the matched words from the search.

The location of the word is important. It is ordered and weighed as per the diagram – e.g. words that match the Summary field will have higher weights than those that appear in the Answer field.

Computed Score

The Computed Score of an answer is usually the same as its Score, unless its Display Position is set to fix it at top/bottom. In that case, the Computed Score is calculated using the score of the answers located at the top or bottom of the list.

To better understand, if a new answer is created, and set with Display Position = “Fixed at the top”, once it is published, its Score will be zero, but the Computed Score will be larger than the highest score for all the published answers.

Score

The Score is a calculated value that ranks the order of answers, and indicates the usage of the answer, as well as how helpful that answer has been to customers. It is calculated based on the Solved Counts:

  • 75% of the score is based on Solved Count, linked to customer usage
  • 25% of the score is based on Solved Count, linked to agent usage

An answer with a large score indicates that several customers (and/or agents) have viewed that answer and that the answer was somehow useful to them.

Solved Count

The Solved Count collects information about the usefulness of answers in the Knowledge Base. Two types of data is gathered:

  • Implicit data – compiled by how customers select and view answers. If a customer views an answer, the solved count of the 1st answer is increased, but not as much as the 2nd viewed answer. In other words, the answer that the customer views last receives the largest solved count increase.
  • Explicit data – compiled by how customers rate individual answers – from the responses to the question “Is this answer helpful?

 

Customer Service: Answers – Organizing answers

Through click-track analysis and feedback on answers, your answers are automatically organized. Answer rankings are constantly updated using the solved count value and presented to your customers with the most useful information first. RightNow uses 3 techniques to gather useful information about answers.

  • Explicit customer feedback: By default, the Answers page offers customers a way to rate answers through the “Was this answer helpful?” option. Their responses automatically raise or lower the solved count of answers.
  • Explicit ranking of the answers: You can explicitly rank answers at certain levels in the knowledge base using the Display Position setting. This is often useful when new issues arise.
  • Click-track analysis: RightNow analyses the path each customer takes through the knowledge base. The use and benefits are two-fold:
  1. Answer relatedness: An affinity map is built which relates answers that customers view to other answers viewed during the same visit. Through the SmartAssistant feature, RightNow Service suggests answers to them based on the historical relationships of that answer to other answers in the knowledge base. The suggested answers appear as learned links on the Relationships tab.
  2. Implicit ranking: The click-track data is also used for answer ranking. Each time an answer is viewed by a customer or suggested by an agent, its solved count is increased

The Solved Count feature collects information about the usefulness of answers in your knowledge base and uses this data to rank your answers. Implicit data is compiled by how customers select and view answers. Explicit data is compiled by how customers rate the effectiveness of individual answers. Both long-term and short-term solved counts are used to calculate the score. Solved counts from the customer portal account for 75% of an answer’s score, and agent solved counts for 25 %. Using the solved count values, RightNow can dynamically rank the answers by their usefulness and present customers with the most effective answers first.

An answer’s score value is a calculated value equal to the answer’s solved count combined with any “fix at” positions specified for the answer in the Display Position drop-down menu when adding or editing an answer.

  • Implicit ratings: gathered as customers view answers. If a customer views an answer, the solved count of the 1st answer is increased, but not as much as the 2nd viewed answer. In other words, the answer that the customer views last receives the largest solved count increase. Previously viewed answers receive a smaller increase in their solved counts. The solved count is also increased when an agent uses a SmartAssistant suggested answer when responding to a customer’s question.
  • Explicit ratings: gathered from the response to the question, “Was this answer helpful?”. This question is displayed on the Answers page on the customer portal.

Over time, an unused answer’s solved count will gradually decline or age. For example if an answer has not been viewed for 30 days (the default setting), the solved count will automatically be reduced. The solved count of unused answers also declines at a constant rate over time. Ultimately, if an answer has not been viewed for an extended period of time, the answer’s solved count can reach zero.

When an answer’s solved count reaches zero, this usually means that the answer has not been viewed for a long time, and it is safe to assume that the information may be outdated or not useful. By default, these aged answers are automatically set to the Review status when their solved count reaches zero. This enables you to easily sort the outdated answers and update them.

When a customer submits comments or suggestions from the feedback form, an unresolved incident is automatically created.