In this paper, we explain how highly unstructured domain knowledge can be acquired and integrated in a case-based reasoning system. We apply our approach to the life counseling domain. We introduce the two steps of our knowledge acquisition approach in such unstructured domains. The rst step is manual and relies on domain experts. The second step is automatic and uses information extraction techniques. Our approach has the potential to contribute to the formalizing and establishing of an important subset of life counseling terminology. In addition, our approach could serve as an example for comparable weak theory domains.
Case-Based Reasoning (CBR) is a methodology for problem solving based on
the fact that previously experienced knowledge can be used to solve new
problems [
Our application SeBaPort (Portal for counseling, in German Seelsorge- und Beratungsportal) deals with life counseling. Life counseling deals with the wellbeing of humans. Life counselors conduct converstions with consulters, give advices and help them to help themselves. Counselors often rely on past expriences for the counseling. The goal of SeBaPort is to help counselors by providing them with counseling cases (depending on their requests), which they can learn from. This makes CBR an ideal methodology to process the knowledge used in that domain.
Life counseling is a domain which is highly unstructured. This makes it very di cult to develop a CBR system for life counseling and be able to provide knowledge in the previously mentioned knowledge containers. For this application domain, we would have to develop an initial set of vocabulary and similarity measures, and also nd a methodology to process the available (unstructured) cases and store them in our case base. SeBaPort does not aim at providing solutions for a given counsel or problem, primarily because the acceptance of the counselors would signi cantly diminish, if we claim to be able to provide complete solutions to counseling problems.
In order to build a life counseling CBR system, we started by developing an initial CBR model and acquiring structured cases. In this paper, we describe in Section 3 how we designed our initial CBR model and our approach for the acquisition of (structured) cases in Section 4. Afterwards we will present some related work in Section 5 and will conclude the paper in Section 6. In the next section, we will rst give an elaborate presentation of the life counseling domain. 2
Life counseling is concerned with the welfare of human beings, more precisely the thinking, feeling, acting, and also the faith of persons. Life counselors help people deal with their problems and con icts. They conduct several counseling interviews with the consulters. The main idea is to help people help themselves by having several discussions with them, give them multiple views on their problem and give them basic hints. Life counselors for example give exercises, which are part of a counseling method, to consulters after an interview. During the following interviews, they try to nd out, whether it helped the consulter or it should be changed.
In order to do that, counselors themselves mainly rely on their experience in the domain, but also on the methodical knowledge they learned during their formation. They are grouped in small communities to share their experiences. As they do not only build on self-made experiences but also on those from others, they often rely on peer consulting and supervision to critically analyse past cases (and be able to learn from them). Further they contact other colleagues when they need help in an actual or past counseling case. Such help might comprise a whole counseling case or just information about parts or aspects (e.g., the method or exercise that can be used in a given situation) of life counseling. Our goal is to provide a system that can be used to help life counselors in their work. We want to provide a decision support system that helps the counselors to document and share their experiences. They would also learn new cases from others and be able to nd hints and references (e.g., to counseling methods) while looking for help (when they deal with a given case). The intended functionality of our system is presented in gure 1 on the basis of the CBR cycle.
An example of the description of the patient's problem in a documented case is given below.
Woman, 48 years old, married and 3 children: 12, 15 and 18 years old. She has been working shifts as a full-time midwife for 2 years. She attends counseling because of insomnia due to her often alternating shift work. She has particularly problems with insomnia after night shifts. She cannot ignore surrounding noises and she cannot completely darken her bedroom. As a consequence of this, she is often tired, is not able to work under pressure and su ers from headaches.
The documentation of the case also contains the documentation of each interview with for example the applied methods, goal validation and solution interventions. 3
When developing CBR systems, one of the rst challenges is to ll the knowledge containers. We have to evaluate which kind of knowledge is availaible in order to know which containers can be lled. In life counseling, the knowledge that is easier to acquire is the experirences made by the experts represented as cases. Due to lack of formalization in the domain, we need to nd a way to extract formalized knowledge from the available cases. For that, we want to structure the information contained in cases. Most experts have their own way to write down their cases. Furthermore, the cases do not contain the same kind of information, have di erent levels of detail and elaborateness. It is thus nearly impossible to automatically detect a structure in raw cases.
Our approach to structure the available cases is to get the needed information from the experts. Instead of getting unstructured cases, we want to be able to get semi-structured cases from the experts. For that purpose, we elaborated a structure that should be used by the experts. The used structure has to re ect the way of thinking of life counselors. We thus have to involve experts in order to de ne such a structure.
Table 2 shows the structure for life counseling cases that we developed. It is
based on a preliminary study done with domain experts (i.e. counselors) [
As another example, the CBR has attributes for the number of children as well as the gender and the age of each children. 4
Now that we de ned a case model, the next challenge is to ll our case base with life counseling cases following the model. Unfortunately, counselors do not have a formal manner to document their cases. This leads to unstructured case descriptions. In order to be able to use those cases in our CBR system, we have to nd methods to formalize the existing knowledge (i.e. cases). This is a di cult task because of the diversity of information available in a case, as can be seen in the last section. Our approch for the knowledge acquisition consists of two steps.
The goal of the rst step is to organize the available information. This is a manual step in which the available diversi ed information is mapped to the structure de ned in Section 3. We rely on domain experts to cope with this assignment. In SeBaPort, this step is realized by providing experts with web forms, which can be used to enter the cases. At the end of this step, we have a case description that matches the structure de ned in Table 2.
The second step of the knowledge acquisition is automatic and consists of using
information extraction to obtain structured CBR cases. The complexity of this
step depends on the type of information given by the experts in the rst step.
Some information, like the gender or the nationality of the patients can be
easily matched to a formal case model. Other, like the medication or the children,
need more e ort for the formalization. For example, the way information about
medication is given di ers from one expert to another and can be more or less
expressive. Nevertheless we have to be able to match the natural language
description of the medication information to our formal model which contains the
additional attributes de ned at the end of Section 3. Another example is the
attribute children, for which the same holds. From the given description, we have
to be able to identify, if given, the number of children, the age and/or gender of
each children and so on. In the example given in Section 2, we would identify 3
children and the ages of the children. However, the gender are not documented.
We used information extraction techniques provided by the component ANNIE
of the framework GATE (see [
Another goal we pursue in SeBaPort, is to be able to learn from the acquired cases in order to formalize the application domain. We thus want to perform a stepwise knowledge formalization for life counseling. This has to be done from scratch because the domain, as explained earlier, is highly unstructured. We are actually trying to gain the formalized knowledge from the acquired cases. The purpose is to be able to tackle the fact that there are not only several ways to document a counseling case, but also di erent counseling perspective. The representatives of each perspective often have problems to deal with case documentations from other perspectives. A formaliztion like the one we are targeting would promote the intercommunication between the representatives of the di ent perspectives. 5
The idea of using CBR in medical related domains has been explored in the
last couple of years. In [
There have been many other CBR applications in medical domains. Nevertheless,
to our knowledge, SeBaPort is the rst one to deal with life counseling.
As for knowledge formalization, there are also other approaches that deal with
that topic. One of them is the knowledge formalization continuum presented in
[
In [
In this paper, we presented the domain of life counseling and how the available knowledge can be used to develop a CBR system (SeBaPort). SeBaPort will help life counselors to extend their knowledge and learn from past cases. We showed how we are actually extracting the knowledge from available cases and we want to use it for the knowledge formalization. We intend to test our knowledge acquisition by evaluating the similarity measures with the acquired cases. The evaluation is still an ongoing work. Furthermore we will develop an approach to incorporate experts' feedback to our formalization process.