University of Arkansas – CSCE Department
CSCE 4613 Artificial Intelligence – Final Report – Fall 2009
Product Sustainability Ontology Project /
Virtual World Ontology Project
Keith Eddy, Matt Hardy, Aaron McGinn
Abstract
In this day and age, making products in a sustainable manner is more important than ever. Data which could assist retailers in this goal exists but is contained within different ontologies. This project consists of first steps toward establishing a method to easily map between these various sources of knowledge so that the retailer can more easily make use of it, then present this information to a consumer within a virtual world such as Second Life. An “ontology engine” will help facilitate this mapping by making use of existing technologies such as Wordnet to automate the mapping the process.
1. Introduction
1.1 Problem
It is desirable for companies to produce their products in a sustainable manner. To this end, various databases have developed to classify products and present sustainability information, such as carbon dioxide released, about these products. However manufacturers and retailers classify products through either UNSPC or GPC which do not keep track of sustainability information. The ontologies that contain the sustainability information do not mirror either of these classification schemes. Therefore, a mapping between the two must be performed. No uniform ontology exists for representing product data and industry categories nor is there a way to augment such an ontology with aspects such as sustainability. Nor is such a resulting “lower level” ontology tied to concept level ontologies.
Virtual worlds do not support an ontology layer of architecture that can be extended to add knowledge to the virtual world. So an object can be labeled “chair” but no additional knowledge can be associated or derived from this label. Expanding on this elementary labeling capability would allow virtual worlds to be more closely linked to the real world. Specifically, Second Life does not support an ontology layer of architecture that can be extended to add real life knowledge to the virtual world. In order to map an ontology to Second Life, we have to be able to somehow display the information. Unfortunately, Second Life lacks direct support for displaying dynamic text through an HUD (Heads up display).
Similarly, reality is not augmented with a knowledge base so we cannot explicitly associate knowledge with places and things. It would be desirable to augment reality with an overlay of additional knowledge. The ontology base in both the real and virtual world can be closely related.
Because there are a limited number of different objects in Second Life that would need a linked ontology structure, a more limited ontology could be created based only upon the labels of objects within. The main problem with creating an ontology for the objects in a virtual world is that the labels added to an object are solely based on the owner. For example, a user might create an object that looks and interacts like a door, but could be labeled “chair.” Therefore the ontology would be incorrect due to the false data
1.2 Objective
The objective of this project is to begin developing a way to create and efficiently maintain mappings between various retail product ontologies and the sustainability ontologies.
As part of this overall goal, there is a perceived need for semantic matching capabilities in order to assist in automating mapping between ontologies. A logical place to start with this in mind is the lexical ontology, Wordnet. Therefore, research must done into Wordnet’s capabilities and operating procedures.
Another objective was to create a working display system using an HUD and link it with an external database/ontology, so when an object in the virtual world was queried, the HUD would retrieve and display the corresponding information or "knowledge" about that object. This information, or at least an object label or name, would have to have been previously entered by someone in Second Life to correspond to the specific object. The ontology would then find the label and return all of the necessary information associated with it, which would be displayed through the HUD.
Goals
· Build a lower ontology (below the level of Wordnet) to account for SKU level product data and product categories and manufacturing processes. Augment above ontology with sustainability attributes and methods to compute a sustainability index. Understand rules of aggregation and inference involved (and record as reasons for decisions).
· Attach product ontology to virtual world Second Life as a way to rapidly populate SL with an ontology layer.
· Demonstrate large-scale augmented reality where the product ontology is available in the real world as it is in the virtual world simulation. This goal is out of scope for present except we can design it on paper with RFID and cell phones plus our ontology..
· Link an external ontology or database to Second Life that can communicate with a HUD and can store information about the objects.
· Create an index of all objects within Second Life along with the frequencies of the object labels and descriptions.
Sub-goals
· Represent ontology information (e.g., RDF/OWL) and edit using an ontology editor (Protégé)
· Represent Wordnet ontology in Protégé
· Attach product ontology to bottom of Wordnet ontology to provide product/SKU level ontology
· Attach sustainability attributes and information into product level ontology (leaving open the idea to add other aspects later, like cost, aesthetics,
· Combine NAICS with …
· Use Second Life search spiders to retrieve named objects. Analyze these and attach to the Wordnet ontology semi-automatically and/or manually.
· Analyze SL ontology data by frequency (many chairs, fewer X-ray machines)
· Analyze SL ontology data by semantic fields (hospital beds are near X-ray machines in a healthcare facility)
· Request ontology information from an object or avatar in Second Life and display or query it
· Create an HUD that can display dynamic text within Second Life
1.3 Context
This project takes place within the long-term Everything is Alive (EiA) project. This project’s aim is to develop technologies and methods to facilitate pervasive computing, the idea that all objects within the world around us can have an identity and can be interacted with through network connections. This would essentially give the real world all of the benefits of a 3d virtual world like Second Life, where anything can be interacted with. Not only does our project plan to extend Second Life giving it access to sustainability information of products represented within it, but it could become the basis of providing a true representation of knowledge of all things within Second Life.
1.4 Potential Impact
Within the next few years, this project could yield improved sustainability data for manufacturers, saving them money and helping preserve our environment. It will also allow for this data to be accessed through Second Life, providing an interactive and intuitive way for consumers to be aware of their own environmental impact. In the longer term, this project could help virtual worlds truly represent our own world by giving objects within them true identity. In addition to scripts, 3d models, and a label, we will be able to attach meaning to an object by relating it to its real world counterpart.
2. Related Work
2.1 Key Technologies
Ontology- For an AI to make decisions about a world, it must have knowledge of it. How this knowledge is represented is its ontology. Traditionally, “ontology” is the study of knowledge. Within computer science, it is the study of how a computer represents knowledge. These computer ontologies are usually hierarchical (a truck is a type of automobile, is a type of self-propelled vehicle, is a wheeled vehicle).
Product Sustainability – Product sustainability refers to manufacturing goods in such a way that it could be continued for and indefinite period of time without degrading the conditions of the natural world or permanently consuming resources. A sustainable product is one whose manufacturing process causes a minimum amount of pollution and uses a high percentage of renewable or recycled resources. For the purposes of this project, we are mainly focused on the emission of carbon during manufacturing.
Virtual Worlds- A virtual world is a 3D representation of some space which can model certain aspects of the real world. The most commonly used virtual world, and the one being used in this project is Second Life.
2.2 Related Work
Linking Lexicons and Ontologies: Mapping WordNet to the Suggested Upper Merged Ontology by Ian Niles and Adam Pease - Describes a project in progress to align SUMO and Wordnet [13]. Discusses such issues as determining the completeness of an ontology. [1]
Ontological mappings of product catalogues - Attempts to formulate easily maintainable mappings between enterprise level product ontologies. But, only addresses general ideas and techniques. [2]
LCA and LCC data semantics sharing across product lifecycle processes - Specifically addresses widespread sharing of lifecycle assessment data (LCA). The paper outlines a model for accomplishing this, but has no full implementation. [3]
Semantic knowledge management to support sustainable product design - Argues for the use of formal ontological and semantic markup languages to allow for efficient sharing of sustainability information. [4]
Life cycle assessment ontology. [5]
A formal approach to product semantics with an application to sustainable design - Proposes a framework with which to accurately and comprehensively describe products with sustainability in mind. Does not deal with mapping already existing ontologies. [6]
Modeling considerations for product ontology - Addresses operational concerns developing a new product ontology. [7]
Practical issues for building a product ontology system - Reports on an attempt to construct an operational product ontology. Addresses how to search an ontology engine among other things. [8]
Design of product ontology architecture for collaborative enterprises - Explores using a generic ontology to facilitate interoperability between enterprises. Includes work on maintaining mappings to cope with updates to the ontologies. [9]
Product configuration knowledge modeling using ontology web language - Presents a way to represent products using OWL and SWRL. Uses a general ontology to define commonalities between various product domains. [10]
Protégé - Protégé is an open source java program that allows the user to create, edit, and navigate ontologies. It provides functionality for helping to merge ontologies limited to allowing the ontologies to be edited side by side and manually mapped. It was our intention to use Protégé as part of the workflow to connect Wordnet to a lower ontology. However, Protégé is incapable of handling the sheer size of Wordnet. [11]
DAML/OWL - a family of languages for representing ontologies developed by DARPA. Programs such as Protégé use OWL as a possible file format. [12]
Wordnet - A lexical ontology that maps the English language. Wordnet uses a hierarchal system to classify words as well as sets of cognitive synonyms (sysnets). It is meant to be a useful tool for computational linguistics and natural language processing research. Over time, several API’s [14] have been created for Wordnet, allowing it to be used in a variety of applications. [13]
WordNet::Similarity - Measuring the Relatedness of Concepts- Describes a fully implemented Perl API for Wordnet specifically meant to facilitate the comparison of words. Implements several proposed algorithms for measure word relatedness. [15]
SUMO - The Suggested Upper Management Ontology attempts to form the ultimate upper level ontology to bridge all ontologies. It is currently mapped to Wordnet but is lacking in lower level implementations that would make it useful for the sustainability project. [17]
In Spring 2009, Anh Chu and Khanh Viet developed an annotation form that enables any user to select an object in Second Life and enter properties about it, which are stored in a remote MySQL database. The user enters commands on a specific channel according to a menu, which is displayed through the chat message box, and can prompt the menu to display properties about specific objects to the chat box (not through a HUD). [18]
Using WordNet-based Context Vectors to Estimate the Semantic Relatedness of Concepts- Describes the use of the vector method to determine if two words within the Wordnet ontology are closely related. [19]
Extended Gloss Overlaps as a Measure of Semantic Relatedness - Describes the “lesk” method for measuring the relatedness of two words within Wordnet. [20]
Creating a search-bot to traverse the worlds of Second Life to amass a database of all objects within the virtual world. [21]
2.3 Related EiA Projects
Our project relates to these other EiA projects, e.g.,
· Mirror Worlds project – our project’s work on ontologies could help objects being mirrored within a mirror world behave appropriately by associating that item with it’s proper attributes and behaviors.
· Search Spider Project- Searching in SL can provide us information on objects (like chairs and castles), the location, ownership, composition, and behavior (scripts) of such objects, objects that occur near other objects (hospital beds near IV drip machines) – and can populate our instance level ontology. Attaching SL objects to the ontology adds a knowledge layer that could be used in deeper searching.
· Soft Controller project – an ontology will eventually be needed to classify the proper actions the soft controller can take with the smart devices being controlled. The API interfaces of a smart device could be stored as part of the device ontology.
· Smart Devices – will eventually be classified into an ontology.
· Chatbots – A chatbot must parse text and make an appropriate reply. Ontologies are important to chatbots because they determine what the chatbot can recognize and respond to.
3. Architecture
3.1 Requirements or Use Cases
· Retailers can access, through some interface, sustainability information on products being manufactured.
o Will require a mapping between NAICS, UNSPC, and the sustainability ontologies.
o Automate this process so that when changes or updates to occur within any of the involved ontologies, the system can automatically update the corresponding mappings
· A model of a real-world retail store with models of real items (a mirror world)
o User can ask the items to display their sustainability information.