Text Analytics processing

This topic provides a conceptual overview of Medallia Text Analytics processing. For detailed configuration information, see Implement Text Analytics.

Discovering and reporting on insights from customer comments is a process that occurs in three phases:

Customer feedback data passes through three phases: text processing, topic analysis and sentiment analysis (parallel processes) and reporting.

During text processing, Medallia Text Analytics performs any necessary machine translations and analyzes the structure of each phrase.

When systems have sentiment enabled, Text Analytics performs both topic analysis and sentiment analysis during the second phase. Topic analysis categorizes each phrase of the text into the appropriate topics, and sentiment analysis detects the sentiment (strongly positive, positive, mixed opinion, negative, strongly negative, or has no opinion) associated with each phrase. 

Finally, Text Analytics populates reports with data, summarizing the actionable information derived from the text and providing you with the ability to investigate further by reading the feedback yourself.

All processing for a records happens between the COMPLETION_PENDING and COMPLETED states of the Survey Status E-field. For more information about Survey Status field values, see Survey status (e_status).

Text processing

During this phase, Text Analytics analyzes the linguistic aspects of the text:

  • Separates the comment into phrases.

    A comment with two phrases. The boundary that separates them is the period between the two sentences

  • Runs machine translations (when applicable).
  • Analyzes each phrase using these techniques: part-of-speech tagging, lemmatization, tokenization, and decompounding.

    If Text Analytics was analyzing the phrase, I had one bad purchase. Analyzing part-of-speech, the system would determine that I is a pronoun, had is a verb, one is a noun, bad is an adjective, and purchase is a noun.

Important: Text processing (also known as parsing) is resource-intensive, and the length of the process depends mainly on the number of comments in the system.

During initial setup, every phrase in the system is processed. For a small system, processing might take less than an hour, while for a much larger system, processing might take days or even weeks.

After Text Analytics is configured and running, incoming phrases are processed on an ongoing basis.

Semantic analysis

A topic finds and captures phrases about a particular subject. You can create topic rules to capture phrases that meet specific criteria. In addition, if you have Topic Finder or Theme Explorer, your system also categorizes phrases through the theme discovery process.

During rule-based topic categorization, Text Analytics compares individual phrases to topic rules. When a phrase meets a rule's requirements, it is tagged by the topic associated with the rule.

  • Phrases that are tagged by a topic are associated with that topic in reports.

  • A single phrase can be tagged by multiple topics. 

  • A single phrase can also meet the same single topic's criteria multiple times, but will only be tagged by that topic once.

For example, in the following image, the phrase meets the criteria outlined for the Issue Resolution and Money back offer topics, so the phrase is tagged by those topics.

A phrase tagged by the Issue Resolution and Money back offer topics

Sentiment analysis

During sentiment analysis (which is available for many industries and languages), Text Analytics examines comments at the phrase level and determines whether each phrase is strongly positive, positive, mixed opinion, negative, strongly negative, or neutral. Using this information, Text Analytics can calculate the percentage of phrases with positive or negative sentiment for any given topic.

The phrase I had one bad purchase has a negative sentiment. The phrase The store was very helpful in solving my problem and getting my money returned to me has a positive sentiment.

For more information about sentiment, see Sentiment analysis.

Reporting

Reports outline the impact of each topic on the organization by highlighting the areas that have the greatest impact on customer loyalty, and by providing actionable information that is essential for identifying problem areas and planning improvements. For example, the following report displays statistics associated with several topics, including the impact of those topics on customer loyalty.

A Topic Snapshot reporting module

For information about how the impact score is calculated, see Impact score. For more information about reports for Text Analytics, see Text Analytics reports.