The Intelligent Systems Group     We invest In the future

Home | People | Projects |Technology | Grants | Collaborations | Associated Centres | Selected Publications | Events | Enquiries

Current Projects

 On Measuring Meaning Similarity between Sentences for Conversational Agents
Who: Zuhair Bandar, David Mclean, James O’Shea, Keeley Crockett, David Pearce, Karen O’Shea
Summary: Conversational agents have usually utilised a simple pattern match approach to search the agent knowledge base to obtain a pre-hand-coded pattern that matches the user utterance. Since "the probability of two people using the same term in describing the same thing is less than 20%", this brings inherent drawbacks to conventional conversational agents. An exhaustive pattern set in a rule is required for a good agent database, compiling such a set is time-consuming and tedious. Moreover this compilation is probably very difficult to be exhaustive. Even if the pattern set is exhaustive, we hardly see much intelligence inside the agent itself. Therefore, a human-like conversational agent with understanding ability is expected to simplify the patterns in agent's knowledge base. This project is towards the development of a method measuring the similarity between meanings of two sentences. The similarity is derived from semantic analysis of main concepts in the sentence. The method is expected to be incorporated with conversational agents, to make them not only to exhibit intelligence on surface but also to process linguistic information like human inside the agent.


 Using neural networks to aid psychological profiling
Who: Zuhair Bandar, David Mclean, James O’Shea, Keeley Crockett
Summary: Visual communication plays an important role in human communication and interaction. Motion analysis offers significant advantages over the analysis of a single image. In this research neural networks are used to extract relevant features from a motion sequence for psychological profiling purposes.

 Conversational Agents Research
Who: Zuhair Bandar, David Mclean, James O’Shea, Keeley Crockett, Majdi Owda,  Karen O’Shea, Annabelle Latham, David Pearce
Summary: Conversational agents (CAs) are computer programs which can participate fully in natural language dialogue with human beings. They essentially allow communication between a user and a computer using natural language.  There are three key methodologies for developing conversational agents. The first and more comprehensive is natural language processing  that studies the constructs and meaning of natural language, applying rules to process information contained within sentences.  The second approach relies on pattern matching of key words and phrases whi8lst the third compares the semantic similarity of phrases to decide what the meaning of the input is (Li, Bandar & McLean 2003), making it more suitable for conversational agents as it will cope with input which is not grammatically correct or complete.   Current project include:

o       Using conversational agents as natural language interfaces to databases

o       Development of an adaptive and self aware conversational agent

o       New methodologies for scripting using sentence similarity measures

 Fuzzy Decision Trees for Classification
Who: Keeley Crockett, Zuhair Bandar, Fathi Gasir
Summary: Decision tree methodologies continue to provide a natural set of tools for extracting patterns from both symbolic and numeric data. Early models such as ID3 have undergone many modifications in an attempt to develop algorithms which can deal with noisy, incomplete and inconsistent information. Recent approaches have involved the use of fuzzy theoretical methods to create "fuzzy trees". By applying such approximate reasoning techniques, more flexible and robust decision trees have been created. The fuzzy tree allows for gradual transitions to exist between attribute values, whilst simultaneously maintaining a degree of transparency in how the decision outcome was reached. Current work is being undertaken on the creation of fuzzy decision trees using a range of hybrid machine learning techniques. This research expands on a novel Fuzzy Inference Algorithm (FIA) which can be extended to act as a multiple classifier.
 
 Fuzzy Regression Trees
Who: Keeley Crockett, Zuhair Bandar, Fathi Gasir (Jay Fowdar – past member)
Summary: Decision Tree induction systems are used for building transparent and highly accurate classification systems. A popular decision tree induction technique used and recognised within the domain of Expert System is, the non-parametric statistical method, Chi-Square Automatic Interaction Detection (CHAID). Recent attempts to improve the efficiency of classification algorithm have involved the use of fuzzy theory to create fuzzy trees, which are able to deal with noisy or incomplete data.

 Natural Language Interfaces to Databases
Who: Majdi Owda, Keeley Crockett, Zuhair Bandar
Summary:    Developing reliable natural language interfaces to relational databases (NLI-RDBs) is a problematic task, since it is related to the ultimate purpose of artificial intelligence, natural language understanding. This paper proposes a new approach for creating conversation-based natural language interfaces to relational databases by combining knowledge trees and goal oriented conversational agents. Goal oriented conversational agents have proven their capability to disambiguate the user’s needs and to converse within a context (i.e. specific domain). Knowledge trees used to overcome the lacking of connectivity between the conversational agent and the relational database, through organizing the domain knowledge in knowledge trees. Knowledge trees also work as a road map for the conversational agent dialogue flow. The proposed framework makes it easier for knowledge engineers to develop a reliable conversation-based NLI-RDB. The developed prototype system shows excellent performance on common queries (i.e. queries extracted from expert by a knowledge engineer). The user will have a friendly interface that can converse with the relational database.

Past Projects

 Using Genetic Algorithms to train Logical Neural Networks.
Who: James O’Shea and Zuhair Bandar
Summary: Logical Neural Networks (LNNs) differ from their connectionist counterparts in that they are composed of logic and memory devices and do not attempt to model structures in the brain closely. During training, patterns from a training set are presented and as learning takes place the RAMs gradually adapt. RAM contents can be saved and manipulated as binary strings, this should make Logical Neural Networks ideal candidates for training using a Genetic Algorithm. This project involves the use of Genetic Algorithms to select near optimum sets of input sub-vectors (inputs to a particular RAM).

 Generalisation in Neural Networks for Continuous Data Domains
Who: David Mclean
Summary: Please see details about the RDSE Algorithm (including a C-coded algorithm and an example of a command file used with the code)

 A Connectionist Approach to Term Extraction
Who: Mr. Peter Marshall
Summary: Following on from previous research into Term Recognition and Extraction, this research will look at developing a connectionist based hybrid learning mechanism to recognise specific medical terms, in context, from medical corpus. Statistic and symbolic based auto-learning methods have already been developed. This will provide the basis of the evaluation.