Application scenarios for semantic functions
All of these wonderful new possibilities which Deep Learning opens up for us must be seamlessly integrated into easy-to-use applications. The different application scenarios therefore require different semantic functions:
Discovery & Monitoring
In order to avoid having to carry out the same activities in the course of ongoing searches, it is possible to automatically carry out recurring searches by means of stored queries and make the search results automatically available in a Dashboard or a Cockpit or in reports (e.g. find everything or find new ones on the subject of XYZ). The relevant metadata and categories for such queries are provided by previous analyses.
In order to open up a large and rather unknown stock of information, navigation is required by means of hierarchical structures. Filter chains (facets) and visualizations of the structures, e.g. in hyperbolic trees, are very helpful. Such structures are difficult to obtain automatically. By means of classification documents can automatically be classified into given structures.
The most common form of research is the spontaneous search for supposedly existing information. Search processes can be very well controlled through analysis results using wizards and recommendation mechanisms (recommendations).
Methodological aspects for gaining of structural knowledge
Previously used vocabulary-based computational linguistics methods are often blind to new aspects in analyzed contents. On the other hand, synonyms, related concepts and conceptual analogies can be learned very well with the help of neural networks and be utilized for the user guidance and suggestion wizards. The task consists of reconciling automatically learned concepts and relationships with known structural knowledge. Automatic procedures must be based on company-specific aspects. There exists no patent recipe for this, but a rather fundamental approach.
The focus is on organizational structure, business processes, topics and, of course, people in the company or working for business partners. Initial basic relationships are already contained in data structures. They are mapped to a system-wide information model when integrating the systems. The organizational structure can also be derived from directory services or the like. Only a few basic concepts which have a structural overlapping effect and which cannot be derived from existing data sources must be explicitly stored in an information model, thus requiring editorial support.
Application example: “Understanding” terms in their context
The use of word models generated by Deep Learning is best illustrated by an example. If a user enters “apple” when searching for information, one cannot draw conclusions as to what the user means by apple. This term can be related to the apple fruit or the corporation Apple Inc.. If you analyze all available information (in the present example, it is magazine articles), you can automatically learn these different conceptual terms and offer the terms actually contained in the existing information to the searcher as a suggestion for the refinement of his search.
In the image, concepts with different contexts are shown in different colors: purple – other inflections of apple, red – terms for apple in the sense of Apple Inc., green – as to apple varieties and so on.
a) Semantic context for “apple”
If the red context is of interest, clicking on one of the red terms takes you to an expanded word cloud (b), which offers even more terms for possible information relating to Apple. The searcher is thus directed automatically to the information available in different contexts, despite a search term which is inadequate for a targeted search. This application of Deep Learning is very easy to integrate into a search and provides great benefit because all the necessary context descriptions are automatically learned.
b) Word cloud for the context of Apple Inc.
As a technique for implementing machine learning, Deep Learning on the basis of neural networkshas great potential for the analysis of digital content. Combined with other analytical methods and the power of today’s hardware, it is increasingly possible to implement economic solutions to support data-driven business processes. intergator by interface projects GmbH from Dresden/ Saxony is such a solution.
Author: Dr. Uwe Crenze is CEO of interface projects GmbH.