Broad capabilities that they wanted in the system were Sentiment on ... insight for product managers, marketing staff an
Solution Scenarios
Confidential
News/Archive Visualization Scenario Problem Exploring large content archives, with emphasis on higher user engagement and content utilization rates.
Business Scenario Searching large archives of digital content has been restricted to keywords, search facets and a linear display of information. For complex unstructured data to be utilized efficiently, information density has to be tuned for effective comprehension. Too much information places cognitive load on users, whereas too little information wastes a user's time.
Key Challenges Look beyond keyword indexing by extracting semantic entities and a map of relationships that connect them. A visual experience that demands minimal user training and aids exploratory analysis and information discovery. An extensible set of “perspectives” that allows distinct view of unstructured text.
Disquery Approach Highlights
An exploratory interface designed for deliberate and sustained analysis. Custom content processing pipeline that uses proprietary algorithms to extract semantic signals of interest. Interface for ad-hoc quick search and formulating complex filter criteria to prune noise and highlighting relevant content. Lenses for additional perspectives. Sentiment Lens, allows grouping of news articles tied to an entity by sentiment. Business Event Lens, highlights articles that contain news on mergers & acquisitions, vendor-supplier relationships and employee job changes.
Confidential
Business Event Tracking Scenario Problem Monitoring news for business events of interest. Simple keyword based alerting mechanisms throw too much false positives to be useful.
Business Scenario A Financial Information Provider wants a new content set to track vendor-supplier events to help analysts monitor their industry landscape. This information would be used to discover broad trends. The provider’s largely manual process to scan and filter daily news articles, press releases and regulatory filings was not scalable. Simple keyword based tracking only increased noise.
Key Challenges
Go beyond keyword tracking, aim for semantic classification of text. Extract semantic entities and business events in a scalable manner from text. Varied formats, many needing custom content ingestion steps and preprocessing steps. Notification of key events in specific industries, or surrounding specific entities, to arrive in near real-time.
Disquery Approach Highlights Tataatsu's Disquery NLP Analytics engine has built-in primitives to help extract entities, subjectivity index, topic classification, sentiment index, monitor and highlight any metric that exceeds configured thresholds. However it was evident that a custom event extractor was required and the solution was build consisting of the following steps.
Creation of event taxonomy capturing primary and derived event variations Custom feature extractors to pull out lexical and semantic signals for specific event taxonomy Custom multi-class classifiers trained on above extracted features Incorporation into Disquery pipeline for identification and notification of flagged events
Confidential
Contextual Collaboration Scenario Problem Enabling organizations with proactive tools and that deliver contextually relevant information.
Business Scenario A global consulting firm wanted to enhance their internal knowledge management and collaboration applications. Goal was to enable employees with information discovery tools that go beyond simple search. Broad capabilities that they wanted in the system were Sentiment on a topic analysis, employee sentiment on strategic/tactical decisions of the company, trending topics and enriched user profiles.
Key Challenges
Creation of semantic map while adhering to content visibility rules. Balancing between deep insights and content ownership boundaries. To utilize contextual signals without intruding upon normal workflow. Deep personalization, highlight content relevant to user’s context.
Disquery Approach Highlights
Scalable content ingestion, semantic signal extraction and mapping, monitoring and notification infrastructure Per-user entity monitoring capability Project-level sentiment tracker, proxy for project health tracking Project level trending terms Contextual annotation and conversation interface Microsoft Outlook and browser plug-ins for seamless information sharing and discovery.
Confidential
Voice of Customer Scenario Problem Automate analyzing customer review and feedback. Highlight insights that feeds into support processes and product management.
Business Scenario Customer reviews for products and subsequent support requests are a key source of critical insight for product managers, marketing staff and support desk professionals. Analyzing this content has traditionally been a manual affair. But growth of social media makes any manual effort prone to delays and errors. Businesses need customer’s perspective to permeate the decision making process.
Key Challenges Customer reviews and product support conversations tend to be of poor quality. Surrounding metadata is minimal. Text tends to be closely tied to domain specific terms. Need to gain insights at a granular level and not just an overall perspective. Need to maintain link between insight and the review that contributed to that insight.
Disquery Approach Highlights
Custom taxonomy that captures product review and support type conversations as two distinct approaches to mapping semantics Custom feature extractors that extract and assign signals to taxonomy created above Custom analytics to capture product feature level sentiment, wished for features and a loyalty metric for customer review data Custom analytics to capture product feature level issues and problems with tool support ecosystem Outlier detection for issues that don’t map to the taxonomy created above And high level trends within existing issue request.
