Data Enhancement
This section describes IDOL functions that allow you to enhance and enrich your data, to provide additional value.
Automatic Language Detection
IDOL can detect the language of a document or piece of text, and use this information to apply appropriate language-specific processing (such as stemming rules, and stop lists).
Categorization
IDOL can automatically categorize your data, by using the concepts found in unstructured text.
IDOL categorization does not rely on rigid rule-based category definitions such as keyword and Boolean operators. Instead, IDOL uses a pattern matching process based on concepts. It can then automatically tag data sets, route content, or alert users to content that is relevant to their user profile.
IDOL hooks into various repositories and data formats, respecting all security and access entitlements.
Category Matching
IDOL can accept a category or piece of content, and return matching categories ranked by conceptual similarity. This ranking determines the most appropriate categories for the content, so that IDOL can subsequently tag, route, or file the content accordingly.
Cluster Information
IDOL can automatically cluster information. Clustering takes a large repository of unstructured data, agents, or profiles and partitions the data to cluster similar information together. Each cluster represents a concept area in the knowledge base, and contains a set of items with common properties.
You can also use dynamic clustering to automatically cluster the results of a query, and then cluster this first set of clusters to create subclusters. This process allows you to generate a hierarchy of clusters to allow users to navigate quickly to their area of interest.
Entity Extraction (Eduction)
IDOL Eduction is a tool that you can use to extract an entity (a word, phrase, or block of information) from text, based on a pattern you define. The pattern might be a dictionary of names such as people or places, or it might describe what the sequence of text looks like without listing it explicitly, such as a telephone number. The entities are contained inside grammar files.
IDOL provides a large set of standard grammars containing useful information that you might want to locate in your content. There are also premium grammars for finding particular types of content, which are more rigorously maintained and updated:
-
PII Grammars. Find personally identifiable information (PII) for a variety of languages and countries, to help you comply with regulations such as the General Data Protection Regulation (GDPR).
-
PHI Grammars. Find Protected Healthcare Information (PHI) in your data, to ensure compliance with regulations such as the Standards of Privacy of Individually Identifiable Health Information implemented as part of the Health Insurance Portability and Accountability Act (HIPAA).
-
PCI Grammars. Find Payment Card Industry (PCI) in your data, to ensure compliance with financial regulations.
-
Government Grammars. Find governmental document markings and other information in your data, to help you comply with data management restrictions.
You can use Eduction separately to find content, and you can use Eduction as part of the IDOL indexing process to extract useful entities and add them to fields for easy retrieval. You can also use Eduction to redact sensitive content that it finds.
Sentiment Analysis
IDOLSentiment Analysis is a part of IDOL Eduction (see Entity Extraction (Eduction)) that can find whether text has positive, negative, or neutral sentiment. For example, you can use it to determine whether users of a particular product or service are satisfied or not, based on an automated analysis of reviews.
The sentiment analysis grammar files contain dictionaries of types of words (such as positive adjectives, negative nouns, and so on), and patterns that describe how to combine these dictionaries into positive and negative phrases.
Rich Media Analytics
IDOL has a variety of tools for Rich Media analytics, which allow you to make the most of your images, videos, and audio files. Rich Media can process media from streams (such as broadcasts, or security cameras providing continuous content) or discrete files, and it can perform analytics such as text capture from images (OCR), face detection and recognition (finding and identifying faces in images), object recognition (such as logo detection), speech-to-text, and speaker identification.
It has a wide variety of applications, including:
-
Broadcast Monitoring. Retrieve and analyze content from ongoing broadcasts to find salient news stories, and process content for analysis. You can use this for many things, from keeping track of developing news stories, to checking how often a brand logo appears in a broadcast.
-
Security and Surveillance. Automate security systems to detect particular events, such as abandoned luggage or traffic infractions, to augment human oversight and reduce human error.
-
Personal Data Protection. Find Personally Identifiable Information (PII) in media such as text in images, faces, or car number plates, and redact it.
Rich Media Analytics is handled by the IDOL Media Server component. The following sections describe the analytics that you can perform with Media Server
-
Optical Character Recognition (OCR). Identify and extract text from images and video frames. Media Server can extract the text from a scanned image of a document, subtitles on video frames, or text that appears on signs in an image. OCR allows you to index the text from images into an IDOL index, so that you can search it alongside your other documents.
-
Face Detection. Detect any faces in an image or video frame, and return the location of each face. Face detection does not require any training, and finds all faces.
-
Face Recognition. Compare a face in an image or video frame to a trained set of faces, and return any matches. Face recognition requires additional training, and returns only matches for your trained faces.
-
Face Demographics. Estimate demographic information from a detected face, such as age, gender, and ethnicity.
-
Image Comparison. Identify which areas of an image have changed, by comparing it to a stored reference image.
-
Image Search (Hash). Search for a particular image in a database of images.
-
Keyframe Extraction. Identify the keyframes in a video. A keyframe is a representative frame following a significant scene change. Keyframes are often used as preview images for video clips.
-
News Segmentation. Analyze news broadcasts to identify the times at which new stories begin and end, and extract the key concepts from each story.
-
Barcode and QR Code Reading. Detect and read a variety of barcodes and QR codes, and return information about the code and the data it contains.
-
2D Object Recognition. Recognize 2D objects such as company logos in documents and images. Object Recognition can recognize the flat image in a scanned document, or it can recognize the image in perspective when it appears in a photograph or video frame (such as a logo appearing at an angle on the side of a building).
-
3D Object Recognition. Recognize 3D objects, such as product packaging in images such as photographs or videos. With sufficient training, Media Server can recognize these objects from any angle.
-
Object Detection. Locate instances of objects that belong to known, pre-defined classes. For example, if you are processing video of a road captured by a CCTV camera, you can configure Media Server to return the locations of all pedestrians, vans, and cars that appear in the video.
-
Image Classification. Classify a whole image as belonging to trained classes, such as 'beach scene' or 'football match'.
-
Color Analysis. Detect the dominant colors in an image or video frame, or in a region of the image or video frame.
-
Clothing Analysis. Find the areas of an image that contain a person's clothing, and return information about the clothing, such as its color.
-
People Counting. Count the people in an image, either by detecting and counting faces, or by using a pre-trained classifier that can identify people in an image (rather than just faces).
-
Automatic Number Plate Recognition. Detect vehicle number plates in an image, in any of the places that these commonly occur on vehicles, and read the number plate values.
-
Vehicle Recognition. Identify the make, model, and color of vehicles that are detected in a scene.
-
Video Event Analysis. Detect events that you consider important in a video, such as cars driving through a red traffic light, people entering restricted areas, and abandoned luggage. Video event analysis can use perspective in the image to calculate speeds and distances. You can set up the system to send alerts when a particular event occurs, or just count how many times a particular event occurs.
-
Audio Categorization. Segment and classify audio into predefined categories such as "speech", "music", "noise", and "silence". You can use audio categorization to inspect an audio file and decide whether to perform further processing. For example, when audio categorization reports that a file contains mostly speech, you might decide to run language identification and speech-to-text.
-
Audio Matching. Identify when known audio clips appear in the ingested media. You can use audio matching to help identify copyright infringement if copyrighted music is played or detect specific advertisements in ingested video.
-
Language Identification. Identify the language of speech in an audio clip or video.
-
Speaker Identification. Identify the people who speak in audio or video, from a trained set of speakers.
-
Speaker Clustering. Find information about the speakers in audio, even if Media Server does not have training data to identify the particular speaker. For example, speaker clustering can show which speaker is speaking in a particular audio segment, and the number of speakers.
-
Speech-to-Text. Extracts speech from the audio and converts it into text. When the audio or video source contains narration or dialogue, you can run Speech-to-Text and index the resulting metadata into an IDOL index so that you can search for topics in the audio, categorize video clips, or cluster together video clips that contain related concepts.
Segmentation and Document Sectioning
IDOL can automatically section your text documents into small chunks so that queries return only the most relevant section or sections.
Summarization
IDOL can accept a piece of content and returns a summary of the information. You can generate different types of summary.
-
Conceptual Summaries. Conceptual summaries contain the most salient concepts of the content.
-
Contextual Summaries. Contextual summaries relate to the context of the original query. They provide the most applicable dynamic summary in the results of a particular query.
-
Quick Summaries. Quick summaries include a few sentences of the result documents.
Taxonomy Generation
IDOL taxonomy generation can automatically understand and create hierarchical contextual taxonomies of information. You can use clustering, or any other conceptual operation, as a seed for the process.
The resulting taxonomy can:
-
Provide insight into specific areas of the information.
-
Provide an overall information landscape.
-
Act as training material for automatic categorization, which then places information into a formally dictated and controlled category hierarchy.