A framework for an agent-mediated knowledge marketplace is presented in this paper. In this marketplace, the buyers are users with questions, and the sellers are users who can answer those questions. The authors use reputation mechanisms to estimate the sellers’ performance, based on their past transactions. Reputation-based markets have inefficiencies, because initially sellers are undervalued until their reputation values come close to their actual ability. The authors suggest that dynamic pricing algorithms can solve these inefficiencies.

The paper defines three kinds of sellers: derivative follower (DF) sellers, who decide on their next bid according to the success of the previous one; reputation follower (RF) sellers, who decide on their next bid according to their reputation; and random (R) sellers, who offer random prices.

The authors ran computer simulations to study agent interactions under two conditions: unemployment, where there is less demand than supply; and over-employment, where a seller will be guaranteed to get a job. Their results show that under unemployment conditions, RF sellers perform slightly better than DF sellers, but under over-employment conditions, DF sellers perform better. In both conditions, R sellers do not behave well, because they cannot adapt to the market.

The main contribution of the paper is to show that while both mechanisms (DF and RF) are successful in helping the marketplace to reach stable conditions, it is important that agents can identify the market conditions and adapt their strategies; the optimal strategy, therefore, maybe a combination of these two strategies.