Key Concepts

Before you learn about testing Text Analytics from the Catalyst console, let’s discuss the Text Analytics features in detail.

Sentiment Analysis

Sentiment Analysis is a common deep learning tool that performs contextual opinion mining of unstructured text, to extract subjective information about the underlying sentiments in it. It involves the applications of general text analysis, natural language processing, computational linguistics, and other machine learning techniques.

The text is generally broken down into its component structures, and weighted sentiment scores are assigned to each entity. This helps the AI determine the overall sentiment of the text, based on a cumulative analysis of the sentiments recognized in each entity.

The Zia Sentiment Analysis model analyzes the polarity of the sentiments in text, and categorizes it as one of these three: positive, negative, neutral.

Zia determines these sentiments for each sentence in a text after analysing them one after the other, and then predicts if the overall text is positive, negative, or neutral, based on the sentiments of each sentence.

Response Formats

Zia Sentiment Analysis returns the analysis of each sentence, as well as an analysis of the overall text as the response. The response also returns the confidence scores for each sentence and the overall text, to showcase the accuracy of the analysis.

The response and confidence score formats differ based on the response types:

• Visual response: The visual response generated when you test Text Analytics in the console displays the sentiment of each sentence in the text, and the overall sentiment of the text in a comprehensive and clear manner. The confidence score of each analysis is presented as a percentage value in the visual response.
• JSON response: You can obtain a JSON response for Text Analytics in both the API and while testing it in the console. The console generates both types of responses. The JSON response also delivers individual sentiments of each sentence and the overall sentiment. However, the confidence score in the JSON response is presented in the range of 0 to 1.

You can view a complete sample JSON response from the API documentation.

Use Cases

Sentiment Analysis is crucial for businesses and brands that place a huge value on understanding customer experiences. Sentiment Analysis can be implemented for the following purposes:

• Gauging and monitoring sentiments expressed by customers in social media platforms and opinion surveys
• Automating the analysis of customer feedbacks and reviews, and reducing the manual workload involved
• Identifying and addressing critical situations in real-time, by the automated monitoring of negative sentiments
• Upholding brand reputation and integrity by delivering quick and thorough customer support, and taking strategic actions
• Implementing a centralized, organized, and unbiased system to arrive at decisions on subjective matters
• Useful for data analysts to conduct nuanced market research and deliver actionable data
• Making informed business decisions and providing tailor-made services based on the analysed sentiments and perceptions

Named Entity Recognition

Named Entity Recognition is a subset of information extraction that identifies named entities in an unstructured text, and classifies them into pre-determined categories. These categories represent real-world objects such as people, places, or organizations. An entity that contains common traits with a category is grouped into it.

NER is concerned with natural language processing and artificial intelligence. It involves undergoing rigorous machine learning using a host of training data to detect and categorize entities with precision. The Zia NER model is efficient and reliable, and can extract and group entities with high accuracy.

Zia NER can recognize and label entities that fit into the following pre-defined categories:

That is, it can determine a word in a piece of text to be the name of an organization or a person, and add it to the appropriate category.

Response Formats

NER returns the recognized entities and the categories they belong to each, along with a confidence score for each categorization that determines the accuracy of the classification.

The response and confidence score formats differ based on the response types:

• Visual response: The visual response generated when you test Text Analytics in the console displays a list of the recognized entities and their categories. The confidence score of each classification is presented as a percentage value in the visual response.
• JSON response: You can obtain a JSON response for Text Analytics in both the API and while testing it in the console. The console generates both types of responses. The JSON response also delivers the entities and their categories, with additional information about the position of the entity in the text. The confidence score is presented as a percentage value here as well.

You can view a complete sample JSON response from the API documentation.

Use Cases

NER is highly useful in identifying key elements in large datasets, and conveying their subject matter by grouping relevant or similar information together. NER can be implemented in the following scenarios:

• Content classification and clustering applications that require efficient, quick, and real-time management of multiple content categories
• Summarizing information in contents like resumes, manuals, news, scientific papers, and group crucial details together
• Customer support applications that categorize complaints or requests based on departments, filter priority words, or pinpoint recurring problems
• Applications in data science and analytics sectors that require extraction and organization of large scale data, or identifying common themes and trends
• Optimizing search and recommendation engines using tags pertaining to entity classifications, and enhancing user experience
• Transforming unstructured data into quantitative and meaningful information

Keyword Extraction

Keyword Extraction is a text analysis technique that involves extracting important and relevant terms from a piece of text, which provides an abstraction of the whole text. It also works on the principles of text mining, information retrieval, and natural language processing.

Keyword Extraction is similar to other Text Analytics feature in the areas of analyzing human language, and developing precision with more training using rich data sets. It uses simple statistical approach like word frequencies and collocations to advanced machine learning approaches.

Zia Keyword Extraction groups the extracted terms into two categories: Keywords and Keyphrases. These highlights deliver a concise summary of the text and provide valuable insights into its topic.

Zia Keyword Extraction returns the response in the same format while testing it in the console, or through the API. Both the visual and JSON response contain an array of Keywords and another array of Keyphrases, which include the extracted terms in them. A confidence score is not returned for this feature.

Use Cases

Keyword Extraction is a highly useful feature if you want to skim through long textual content and just obtain the essential information and action items from it. It enables you to identify the subject matter and the gist of the text at a glance, and save valuable time.

Keyword Extraction can be used for the following purposes:

• Implementation in online libraries that return key highlights of reading materials as search results
• Automating data retrieval, indexing, and organization in large datasets like online forums, customer feedbacks, news reports, and more
• Querying for the presence of particular keywords and keyphrases from extracted material, and performing necessary actions accordingly
• Implementation in applications that perform real-time analysis and generate automated responses based on the extracted key terms
• Obtaining quick and in-depth insights from a huge number of source materials, and saving efforts and time in manual research and processing
• Reducing inconsistencies in manual information retrieval, since Keyword Extraction functions on pre-defined parameters