-Web portals provide a standardized way of integrating multiple information sources and applications in a single web interface. However, they currently do not provide semantic support for users that need to navigate the often overwhelming amount of content. We demonstrate our open source portal architecture “hanu¨ wa” that integrates text mining web services, based on the Semantic Assistants framework, with the Liferay portal server.
INTRODUCTION
Web portals are a specific kind of web-based systems that provide for an integration of diverse information sources and applications. Deployed for a concrete scenario in an organization, they typically address the information needs of a wide range of users and their tasks through both internal and external services.
While a web portal provides convenient access to information, there is no standardized way that allows to further process the available content in order to support users in their tasks. There is also a lack of appropriate technologies for document filtering within a web portal. We envision a new generation of web portals that can provide context-sensitive support through semantic analysis services, in particular based on natural language processing (NLP). These services are deployed in shared or private servers and can be dynamically requested by users that ask for help in a specific task: e.g., finding entities in a documents, summarizing a text, answering a question, or linking content to external sources. As such, they perform the role of AI “assistants” that support their users. Furthermore, we imagine enhancing web portals with a personalization component to adapt the content to the user’s needs. Sorting documents or highlighting terms according to a specific user interests would be a great advantage for the user and a step towards working against information overload.
In previous work, Bakalov et al. [
Our new portlets can be deployed in any existing Liferaybased portal to offer natural language processing services to its users. Here, we demonstrate the core functionality with 1IBM WebSphere, http://www.ibm.com/software/websphere 2Liferay, http://www.liferay.com/ named entity recognition in a given article, but the framework is not limited to a single domain: A clear separation of concerns allows a language engineer to make new NLP services available without requiring knowledge in portal technology, and a web engineer can easily design a new web portal that incorporates language technology.
II.
Our novel Semantic Assistants-portal integration architecture, illustrated in Fig. 1, is designed to allow various portlets to benefit from NLP techniques on their content. The core idea is to enable generic portlets to communicate with the Semantic Assistants portlet, specifically designed to connect to the back-end Semantic Assistants server and provide inquiry and invoking capability of NLP pipelines to portal users.
Portal
Database
Portlet Controller ... POothrtelert Semantic Assistants Server
NLP Service Connector
Service Invocation
Service Information Semantic Assistants
User (Embedded) Browser Language Service Descriptions
In this architecture, all available portlets in a page can communicate with the Semantic Assistants portlet by sending content for analysis and receiving the results. To commence an analysis session, users interact with the portal via their web browser, for example, on their desktop computer or from a mobile device. Through this integration, users can select an NLP service to execute on a portlet’s content from a dynamicallygenerated list of available assistants in the Semantic Assistants server repository. Where applicable, users can also customize the services’ behaviour by setting runtime parameters. An execution request is then sent to the Semantic Assistants server from the Semantic Assistants portlet in form of a W3C3 standard web service call that triggers the execution of the designated NLP pipeline on the provided content. The results of each
NLP Service Results successful service execution are first received by the Semantic Assistants portlet and then passed on to the portlet that requested the service execution. The NLP pipelines are described in the OWL4 language and the Semantic Assistants server uses SPARQL5 for a dynamic discovery of available services upon each user request. Hence, adding or removing NLP services to the integration requires no modification to the code base of the portal.
The basis of the personalization component will be an
ontology-based user profile, where all user interests are recorded
automatically while browsing through the portal and reading
documents. A user interface, embedded into a portlet, allows
a user to control interests, add new terms, delete or change
concepts. The user can also enable or disable the personalization
mode. When personalization is desired, the documents are
resorted and the relevant terms of the user profile are highlighted
within the text. In contrast to [
III.
APPLICATION
The integration of NLP assistants within a portal context allows for a multitude of applications. Fig. 2 shows an example scenario in which a portal user needs assistance in analyzing the textual content available in the content portlet (left). Such assistance can be offered to the user through the NLP services listed in the Semantic Assistants portlet (right). This portlet allows the user to connect to different Semantic Assistants servers and review the list of their available pipelines in order to find a suitable assistant for his task at hand. In our example, the list of assistants contains a “Person and Location Extractor” service that extracts entities of person and location types from a given text. The user then sends the text in the content portlet to the Semantic Assistants portlet for analysis and requests the service execution by clicking on the “Run Assistant” button. This interaction will request the designated Semantic Assistants server for the execution of the ANNIE pipeline, provided by GATE.6 Subsequently, the results are returned to the content portlet in form of annotations in a tabular format and highlighted in the text based on their offsets. The processing time for different scenarios depends on both the length of the input text and the actual NLP pipeline. Naturally, sophisticated NLP pipelines with deep syntactic or semantic analysis require more time to process. Currently, we are working on a personalization scenario aimed at tackling the user’s information overload issue, by filtering the portal’s content according to a user’s interest. The idea is to embed such capability directly within portlets, allowing users to be able to switch to various personalization modes.
IV.
In this paper, we described our open source integration of natural language processing capabilities within a portal environment. We also intend to integrate a personalization feature into portals to adapt their content according to a user’s needs. Furthermore, we want to provide a user interface to give the users the opportunity to have control over their recorded interests. The NLP-portal integration will be available as part of the Semantic Assistants distribution hosted on SourceForge.7