A semantic wiki for novelty search on documents

Michael Färber, Achim RettingerDIR 2013 · OpenReview

Abstract

Technology-driven organizations face significant challenges in "technology watch"—the process of monitoring emerging trends and inventions. Standard search engines often fail to distinguish between generally relevant information and information that is truly novel to a specific organization. This paper presents a semi-automatic system that integrates Semantic MediaWiki with ontology-based information extraction to identify and rank documents based on their relative novelty to an existing internal knowledge base.

Introduction

For technology-oriented companies, staying ahead of market shifts and new patents is a critical but labor-intensive task. The primary obstacles are information overload and the "novelty gap." While standard search tools can find documents relevant to a keyword, they lack context regarding what a company already knows. Consequently, experts spend excessive time filtering through redundant information. This research aims to bridge that gap by creating a system that understands the user's current state of knowledge and specifically targets the "missing pieces" in their domain ontology.

System Architecture and Workflow

The proposed solution centers on a semi-automatic loop utilizing Ontology-Based Information Extraction (OBIE). The workflow, depicted in Figure 1, bridges the gap between unstructured web data and a structured internal Knowledge Base (KB).

The semi-automatic workflow illustrating the loop between the internal knowledge base, web search, and fact extraction.

The process involves four key stages:

  1. Contextual Query Generation: The system analyzes existing wiki entities to create structured search queries based on known relationships.
  2. Semantic Ranking: Search results from databases like Espacenet are converted into semantic graphs using Named Entity Recognition (NER) tools like GATE and Wikipedia Miner.
  3. Novelty Detection: The system identifies documents containing entities or relationships absent from the current KB, ranking them higher than documents that merely repeat known facts.
  4. Human-in-the-Loop Population: Domain experts review and import novel facts, providing feedback that refines future search weights.

The Concept of Relative Novelty

A fundamental contribution of this work is the definition of novelty as a relative property. By representing knowledge as a graph, the system can mathematically quantify novelty as the "delta" (difference) between the document's semantic representation and the organization's internal graph. This allows the system to search "out of context"—for example, looking specifically for properties of a technology that are not yet recorded in the company's database.

Prototype Implementation

The authors developed a functional prototype where users can manage technology profiles within a Semantic MediaWiki environment. As shown in Figure 2, the interface allows experts to interact with extracted data directly.

The prototype GUI allowing experts to review and semi-automatically populate the knowledge base from extracted facts.

In this example, a user viewing a "Lithium-ion battery" entry can see current properties and use an edit form to import new data found via the novelty search. This integration ensures that the knowledge base remains a living document, updated with minimal manual effort. The system learns from these interactions; if a user frequently imports a specific type of information, the system increases the weight of those relations in future rankings.

Limitations and Future Work

The system's performance is currently bounded by the accuracy of underlying NER tools. Furthermore, comparing complex semantic graphs across millions of documents presents significant computational challenges. Future research will focus on optimizing these graph comparisons and conducting formal user studies to evaluate the system's impact on expert productivity in real-world technology watch scenarios.

Summary updated

BibTeX
@inproceedings{farber2013semantic,
  author = {Michael Färber and Achim Rettinger},
  title = {{A semantic wiki for novelty search on documents}},
  booktitle = {DIR},
  year = {2013},
  url = {https://openreview.net/forum?id=cBCMQm-thx},
}
Figure