SAS^® Sentiment Analysis

Features

Statistics and linguistics combined to provide more accurate sentiment analysis results

• Provides a choice of approaches to sentiment analysis:
  • Statistical modeling: Provides predefined default parameters, as well as the ability to modify parameter settings to identify the key elements for sentiment evaluation from textual collections.
  • Linguistic rules: Lets analysts define the elements to be examined for sentiment reviews through contextual Boolean operators.
  • Hybrid approach: Provides the unique ability to use both statistical rigor and business expertise to define elements and features and how they should be combined for the most accurate sentiment evaluations.
• Ability to import and/or create concepts for evaluations.

Context of features examined for accurate interpretations

• Assigns weights to matches to determine the optimal concept match for any particular document or source.
• Supports complex linguistic rules for one or several matches of a term, regular expressions, co-referencing, part-of-speech tags, the distance or occurrence of concepts in relation to other words, and more.
• Includes prebuilt tasks to simplify linguistic pattern identification. It offers:
  • Predicate rules to define semantic relationships between concepts.
  • Operators to locate related information with greater precision, such as co-referencing.
  • Identification of intermediate concepts that contain rules referenced by other concepts (shortens the rule-writing process).
  • Case-insensitive matching, so you can match both uppercase and lowercase terms.
  • An operator to reference the polarity in keywords.
  • Easy modification and reuse of existing rules with cut, copy, paste and other quick keys.
  • Rule editing within the body of the rule.
• Checks rule syntax after editing, before the rule is built, to improve rule construction.
• Offers a search facility to find keywords within rules.

Dynamic sentiment analysis

• Use phrases to express connotation with combinations of Boolean rules and other linguistic tools.
• Permits improvements and/or changes over time with multiple model generation within the same project, and the ability to edit and test model modifications.
• Allows subject-matter experts to refine model performance with an easy-to-use, point-and-click interface.
• Monitoring of results over time will inform model analysis refinement, and comparisons to benchmark training sets can be established to allow for continuous improvement.

Easy-to-use interface

• Project wizard allows you to configure your project while it is being defined.
• Specific definition for statistical model and rule-based model configurations.
• One view of model test and configuration settings provides greater clarity.
• Panes in the interface simplify tasks and display more information, such as rule evaluations and search results.

Interactive workbench

• Ability to add new concepts/entities to capture desired topics.
• Word cloud report, based on defined concepts, automatically illustrates extracted noun phrases.
• Prebuilt operators can be used to hone keyword searches.
• Online help documentation is integrated with the interface.
• Some workbench functionality is available through Web services APIs so documents can be programmatically added to different projects, exported, managed, searched, etc.

Web-based reports

• Standard output reports include:
  • Visual sentiment trend reports indicate the positive/negative and neutral classification frequency of defined concepts over time.
  • Overall sentiment rating report describes the overall sentiment as well as those associated with all of the defined features.
  • Ability to explore results in more detail with SAS Business Intelligence tools and data visualization and discovery capabilities with add-on capabilities.

Updates on Web postings, reviews and opinions

• High-performance, multithreaded crawler that can be deployed in a distributed mode to maximize processing and support extremely large-scale crawls.
• Powerful linguistic technology is built in to extract URLs from Java scripts.
• Web crawls can be interrupted and resumed.
• Web crawler lets you:
  • Define individual quotas for a particular Web address.
  • Add a field with a constant value to each input document.
• Document processor makes it easy to parse, export and control processing and output.
• Duplications are automatically removed.
• Politeness requests help prevent complaints or blocking of access by enabling a friendly download feature.
• Supports desired restrictions on crawling to customize searches (e.g., to specific file formats, servers, restrict the area or specify search depth).
• Crawls can be set to incrementally update only changes to previous saves.

Multiple languages supported

• Full suite of 27 languages (plus dialects) available.
• Supports Arabic, Chinese (both Simplified and Traditional), Czech, Danish, Dutch, English (US/UK), Finnish, French (French/Canadian), German (New/Old), Greek, Hebrew, Hungarian, Indonesian, Italian, Japanese, Korean, Norwegian (Nynorsk/Bokmål), Polish and Portuguese (Portugal/Brazil), Romanian, Russian, Slovak, Spanish (Spain/South America), Swedish, Thai, Turkish and Vietnamese.