Facilitating Dialogue - Using Semantic Web Technology for eParticipation

Facilitating Dialogue - Using Semantic Web Technology for eParticipation George Anadiotis, Panos Alexopoulos, Konstantinos Mpaslis, Aristotelis Zosakis, Konstantinos Kafentzis, and Konstantinos Kotis IMC Technologies S.A. 02/06/2010

Presentation Structure Introduction: on eParticipation Adding semantics to eDeliberation Finding relevant views Sharing views Evaluation Conclusions and future work

eDialogos: our approach to eParticipation eParticipation: ‘the use of ICT to broaden and deepen political participation by enabling citizens to connect with one another and with their elected representatives’ Our approach is multi-disciplinary, combining elements of political science and communication, software and knowledge engineering to develop: An intuitive and non-disruptive methodology for efficient and transparent distributed decision-making A solid technical foundation that provides advanced functionality via a seamless user experience, based on Semantic Web technology while hiding its complexity behind a user-friendly Web2.0 interface.

eParticipation methodology ePetition: bottom-up, reflects well-known petitioning process eConsultation: top-down, collects feedback from citizens eDeliberation close collaboration between decision makers and citizens to formulate policies and achieve consensus tight 'serial process' within a specific time-frame

Using ontologies to model eDeliberation domains Each deliberation process is focused on a specific domain, determined by the outcome of the agenda setting stage modeled as a community that stakeholders can join and gives access to a set of tools and functionality to promote information sharing, views expression/quantification and policy formation For every domain, an ontology can be created/reused and associated with the deliberation community users are given context for the topic under discussion, in a form they can use to annotate and browse content based on provided annotation and ontological relations, a powerful hybrid search mechanism is implemented

Case Study: Prefecture of Samos http://www.samos-dialogos.gr Domains Recycling Culture Tourism Ontologies created by the University of Aegean, AI-Lab, using collaborative authoring methods and tools (HCOME, HCONE2) manual document processing related ontologies Internationalization: artifacts named in English, multilingual labels Ontologies are processed to ensure appropriate labels and stored in a repository

Using ontologies, the Web 2.0 way: Associating Administrators may browse and select an ontology to create a 1-1 association with an ongoing eDeliberation The selected ontology is automatically imported and processed to create a ‘taxonomical equivalent’, retaining only classes, instances and their hierarchical relations, using labels as taxonomy terms.

Using ontologies, the Web 2.0 way: Annotating The resulting taxonomy can be used to annotate items within the eDeliberation community space in the form of tagging

Using ontologies, the Web 2.0 way: Browsing Annotations are utilized for browsing, in the well-known form of tag clouds (overall and per eDeliberation)

Search mechanism requirements: the need for a hybrid search mechanism In addition to browsing, we wanted to implement a search mechanism appropriate for our context, thus being able to perform well in the face of the following requirements Usability by general public. Involving as broad an audience as possible -> intuitive search mechanism Applicability to heterogeneous content. Variety of tools used to support the deliberation process, each one generating content with different characteristics -> combination of full-text search and user-provided annotation Applicability to user-generated content. Not all resources will be properly annotated -> combination of full-text search and user-provided annotation

A hybrid search mechanism: full-text search + ontological annotation Each piece of content generated by the platform tools is treated as a document and indexed using Lucene Title, tags and actual content are linguistically analyzed (tokenization and lemmatization), then indexed as discrete fields Query string is parsed using the same analyzer used for indexing, ensuring that query terms match index terms and the resulting terms are looked up in the index Ranking results is done using standard TF/IDF metrics, however the query performed against the index combines field scores giving different weights to each Annotation ( = Tags) 1st, Title 2nd, content 3rd and custom content fields (where present) follow

Query refinement Query refinement works on the deliberation level If users choose to use the search functionality while browsing a specific deliberation space, they can restrict results to content generated only for the deliberation at hand.

Query expansion For a given query term, find terms that are semantically similar in the specific search context. Based on Alexopoulos et. al. ‘A fuzzy ontology framework for customized assessment of semantic similarity’

Query expansion Combines domain ontologies with fuzzy logic Based on a context model that captures information about which of the domain ontology's relations and to what extent should participate in the similarity assessment process Any ontological relation can be used Parameters are predetermined, used by a contextualization algorithm that produces a term semantic similarity index, subsequently used for the expansion

Query expansion sequence Query parsing. Query terms are parsed and compared against taxonomical terms -> list of taxonomical terms. Query expansion. Taxonomical terms are related to ontological terms via URIs. The list of URIs that correspond to the taxonomical terms is sent to the query expansion component, which then uses the semantic similarity index in order to produce a list of related ontological terms. Query execution. The list of related ontological terms is associated to their taxonomical counterparts (via URIs) and used as input to the hybrid search mechanism. The weight of terms retrieved via the query expansion mechanism is appropriately reduced.

The need to share For any given discussion topic, getting a holistic view involves knowing what others think on the topic, background information, existing arguments etc These are distributed over a wide range of resources: Fora, social networking sites, mailing lists, newsgroups

Linked Data as the sharing mechanism We choose to enable sharing directly on the data level, via Linked Data: Standard, increasing adoption Specify the semantics of the exchange (ontologies) Provide a mechanism to support identification and cross-reference of resources (URIs) a transport layer and remote structured querying facilities (HTTP, SPARQL) Consume and produce (Inbound/Outbound Approach)

Outbound: eDeliberation Ontology Makes the structure and semantics of the framework explicit, relying on standard ontologies (SIOC, MOAT, Tagging Ontology) Makes content accessible to external applications as Linked Data

Outbound: Mapping and providing Linked Data via D2R Server Mapping the platform’s relational database to the selected vocabularies using D2R Server + mapping language RDF available via SPARQL endpoint

Inbound: Distributed contextual view retrieval Aims to give an insight on what has been discussed on any given topic in other fora: retrieve related views. Can use other eDialogos instances or any Linked Data-aware discussion forum as input Granularity: Query performed at the thread level Generic enough to be able to query any SIOC-aware forum Utilizes extra deliberation-specific information if present Similarity: using a mixture of structural information and content similarity on the textual level No assumptions about the presence of annotations Generic enough to be able to take such information into account, if present

Quantitative Evaluation: Access Statistics Prefecture of Samos facts: 33.814 people 42% of the population are internet users, out of which 60% have broadband access After 1 month of use: Number of citizens who have registered at the platform: 70. Number of citizens who participated in the first agenda setting e-poll: 25. Number of citizens who posted to the e-forum: 20 with 33 posts. Number of e-surveys submitted: 4 answered in total by 44 citizens. Monthly average statistics: 784 visits, 385 unique visitors, 6.469 page views. Assuming that broadband users are the ones most likely to use such a sophisticated platform, then we can -very roughly- estimate that: 1% of them became registered users 10% participated in one way or another

Qualitative Evaluation: User Feedback

Future Work Improved support for annotation : semi-automated annotation based on content/ domain ontology/context (deliberation) Meanings for user-provided tags : taxonomical terms are bound to their ontological counterparts via MOAT. User-provided tags may also have meanings (pending Greek DBpedia) Inference : 'a certain amount of inference (datatypes, basic RDFS, owl:sameAs) would be desirable and feasible‘ for D2R SPARQL endpoint Refined similarity measure for distributed contextual views retrieval : Semantic IR (pure SPARQL insufficient) Incorporate ontology matching techniques Argumentation theory: add structured dialogue features

Conclusions Using Semantic Web Technology can improve functionality and usability, if the complexity is hidden from end-users The use of Linked Data particularly for the eParticipation domain is an ideal match, as it promotes both interoperability and transparency Going one step further: Open-sourcing eDialogos to maximize adoption, promote dialogue ecosystem

Facilitating Dialogue - Using Semantic Web Technology for eParticipation

More Related Content

What's hot (18)

Similar to Facilitating Dialogue - Using Semantic Web Technology for eParticipation (20)

Recently uploaded (20)

Facilitating Dialogue - Using Semantic Web Technology for eParticipation