Adaptive genetic algorithm to improve negotiation process by agents e-commerce
In the past decade we have witnessed the rapid growth of negotiation agent systems in electronic commerce (e-commerce). A huge number of works have been done in order to develop adaptive negotiation agent systems. However, existing models are weak in supporting real world e-commerce conditions since...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2011
|
| Subjects: | |
| Online Access: | http://psasir.upm.edu.my/id/eprint/27378/1/FSKTM%202011%2017R.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In the past decade we have witnessed the rapid growth of negotiation agent systems in electronic commerce (e-commerce). A huge number of works have been done in order to develop adaptive negotiation agent systems. However, existing models are weak in supporting real world e-commerce conditions since most of them assume static conditions for environment and complete information about opponent agent’s preferences. E-commerce platforms are dynamic and flexible environments which require adaptive negotiation models to survive. In this situation, non-adaptability of models results into non-optimal or sub-optimal performance which is an obvious problem for previous models. One-to-many negotiations are the most common types of negotiations in e-commerce platforms. Also, many-to-many negotiations are a form of parallel one-to-many negotiations. Therefore, in this study we focus on improving one-to-many negotiation model, so that later we will be able to expand it to many-to-many negotiation. E-CN one-to-many negotiation model has been chosen as our benchmark. E-CN model is a popular one-to-many negotiation model which is currently being used by British Telecommunication. This model has later been improved by ODC. This study is using the same data set and performance measurements explained in e-CN model. The objective of this study is to improve an adaptable version of e-CN model which is able to adapt according to the opponents’ preferences and environmental condition while tackling with local optimum answers. This adaptability property improves the performance of the proposed one-to-many negotiation model by producing high quality and mutually acceptable offers. The proposed adaptive negotiation model is named Aspirated Genetic Algorithm (AGA) negotiation model which is a hybrid negotiation system composed of different negotiation strategies, Aspiration concept,genetic algorithm and Bayesian learning. The proposed adaptive AGA negotiation model is a hybrid negotiation system that is composed of different negotiation strategies, Aspiration concept, genetic algorithm and Bayesian learning. In order to reach a high quality solution and guarantee the success of negotiation model two contributions have been made. The first contribution is to improve the decision policy of e-CN negotiation model using Aspiration concept. This contribution is made in order to tackle the local optimum offers and heightening the risky attitude of agents. Second contribution is made by applying an improved genetic algorithm in offer generation mechanism and producing high quality, mutually acceptable offers. This improved genetic algorithm considers an economic encounter where both parties (sellers and buyer) preferences are considered when generating new offers. The proposed negotiation algorithm employs Bayesian learning and similarity functions in order to predict opponent agent’s type and preferences. The results of this study showed that the proposed negotiation model improved agent’s learning ability and decision policy in order to adapt to the changes of the environment and tackle the local optimum answers. In addition, proposed negotiation model generated mutual acceptable offers that reduced the number of negotiation rounds before reaching agreement; this resulted in faster agreement while reducing the negotiation cost in comparison with ODC model. The proposed negotiation model has been studied over different dynamic conditions using the same data set and measurement metrics as e-CN. Experimental studies demonstrate that the final utility value, joint utility value, negotiation time and success rate of proposed negotiation model has significantly improved in comparison with e-CN model as our original benchmark and with ODC as a recent negotiation model. Finally, this thesis provided an improved adaptable negotiation model which guarantees reaching agreements while supporting the real world B2B scenarios in e-commerce. Further studies needs to be undertaken in order to evaluate qualitative issues as well as quantitative issues. Also, there is an open direction to accelerate the speed of proposed genetic algorithm in order to reduce the cost and time of negotiation. Also a possible further study is to evaluate the performance of this model on a many-tomany system architecture. |
|---|