The Affective Guide possesses a long-term memory that is made up of declarative memories, both semantic and emotional. These memories contain a complete set of information when the interaction between the user and the guide starts. In contrast, the user interests model is empty. The guide makes assumptions about the user’s interests based on the initial information from the ice-breaking session. Since a life story is always more interesting than simply bare facts, we argue that an emotional memory is important and necessary for the guide. The emotional recollection of past experiences will allow the guide to tell more believable and interesting stories. In addition to the long term memory, the guide's current memory holds information relating to recent processing.

The guide defines two informal ontologies. First, a story element attributes ontology, which consists of the attributes used to annotate the story elements. These attributes and their corresponding interest areas - General, Science, Military, Politics and Social are hierarchically structured. Second, the guide profile ontology describes the guide's role and interests. Besides these ontologies, the system contains definitions for basic entities, including event, concepts, personnel and divisions. These definitions are used to introduce related events, concepts, personnel and `Los Alamos' divisions respectively, the first time they appear in the narrative process. These entities are tags in the story elements that serve as triggers to the remembering process during story activation and extension.

Figure 1. The Narrative Construction Process

After the spot has been selected, the guide proceeds to extend it, step 2 in Figure 2. Given that one person can remind us of another person, one object can remind us of another object or one event can remind us of another event, story extension is activated based on subject-object links and cause-effect links. Moreover, the guide uses other entities such as story type, concepts, interest attributes and location as activation factors. The story elements are organised using a network-like structure so that associative activation can be performed during extension. All extensions are performed by JESS, a Java based rule engine to perform reasoning. When the desired granularity is reached and the combination of the story elements is large enough to generate a short story, the extension process is complete.

It is noteworthy that the guide includes its ideological perspectives only if it is currently competent and highly certain of the user's interests, that is, when user shows interest in the stories or agrees with its argument. The EEs are selected based on the activated SEs to ensure a smooth flow of storyline. To reduce complexity in the current version, a SE will lead to inclusion of only one EE. Hence, the number of EEs added is always less than or equal to the number of SEs.

Each EE usually follows its associated SE. However, if two SEs have a cause-effect relationship, then the associated EEs will come after both SEs. Next, all the selected elements are combined taking into consideration the existence of entities. The guide retrieves the definition for each entity, (event, concept, personnel and division) that appears for the first time whether in SEs or EEs. The resulting story is then send to the PDA to be presented to the user.

After each story presentation, the guide updates its current memory so that it can be reminded of the current active memory elements in the next retrieval cycle. Analogous to human memory, a concept strength in the guide's memory increases when it is activated frequently and will be forgotten if not used after a few iterations. In contrast, the current activation of personnel, attribute and type remain in the current memory only for a single story cycle to allow flexibility in story generation.

           

Figure 2. The Spot Selection and Story Extension Process