Readings – Text Analytics: Practical Implications and Applications
What Text Analytics Can Do for Us Today
A couple of weeks ago, one of my DS 453 Text Analytics students asked me: What are the practical uses of text analytics?
Fortunately, around the same timeframe, another student (same class) emailed me a link to a really great paper showing a superb application of pretty straightforward text analytics:
Good, Strong Text Analytics Application
- An Advanced Analytics Approach for Recall in Automotive Industry – pdf; link courtesy L.K., Cohort 11 (Winter 2018), Text Analytics class at Northwestern University School of Professional Studies Master of Science in Data Science program. This is an important and useful paper, because it shows how using fairly straightforward text analytics methods (entity extraction), there can be a very strong influence on industry practice. This method works so well, in part, because when dealing with automobile makes, models, and manufacturers, the extracted entities tend to be precise; there is little ambiguity in the match-up. It is also fairly easy to extract what the nature of the customer complaint is. These two aspects, together, make for a pretty good text analytics application.
Signal-to-Symbol Representation in AI
- Symbol Emergence in Cognitive Developmental Systems: a Survey – I keep reinforcing how a core notion in AI is that of representation levels, and a lot of the work is to move from signal through syntactic / semiotic up to symbolic. This paper addresses how the symbolic representation level can emerge.
- Query Focused Abstractive Summarization: Incorporating Query Relevance, Multi-Document Coverage, and Summary Length Constraints into seq2seq Models – this paper deals with summarizing abstracts, but summarization techniques are overall important – they can work on news sources, etc.
- What did you Mention? A Large Scale Mention Detection Benchmark for Spoken and Written Text – this works with entity co-resolution, which is matching extracted entities to a base set of known entities; very important task!
Sentiment Analysis
Sentiment analysis is one of the trickiest things in text analysis, for many reasons.
Here are a few interesting papers, drawn from arXiv-sanity’s listing for late January, 2018:
Ontologies (Knowledge Graphs) and Text Analytics:
- Knowledge Graph Embedding with Multiple Relation Projections – connecting knowledge graphs w/ text
- Etymo: A New Discovery Engine for AI Research – “Etymo constructs and maintains an adaptive similarity-based network of research papers as an all-purpose knowledge graph for ranking, recommendation, and visualization.”
- A Question-Focused Multi-Factor Attention Network for Question Answering – answering questions based on text is complex; this makes an effort to create more robust answers
- DKN: Deep Knowledge-Aware Network for News Recommendation – important topic and application.
Connections between Image Understanding and Text, Including Captioning
- Improving Bi-directional Generation between Different Modalities with Variational Autoencoders – connecting text with images
NOT Text Analytics; stashing some links here for use when I teach Machine Learning in Spring, 2018:
This one’s for me, for later on – just didn’t want to lose track of it:
- Discovering Markov Blanket from Multiple interventional Datasets – does this relate to what I’m doing in working with Karl Friston’s notion of a Markov Blanket, and my notion for creating a representation using the 2-D CVM as an internal computational engine?
- Using KL-divergence to focus Deep Visual Explanation – just because I think it might be good
- CommanderSong: A Systematic Approach for Practical Adversarial Voice Recognition – oh, this is scary stuff!
- Multi-optional Many-sorted Past Present Future structures and its description – dealing with temporal organization of events
- DVQA: Understanding Data Visualizations via Question Answering – Q&A about bar charts in documents
Follow-up for me: