站着说话,不嫌腰痛 。Because I am not a professor, I cannot have any students. All I write here is my "day dream," WHAT IF I were a professor and had funding to support one Ph.D student. How should I pick this student from more than five applicants? I will take each recommendation letter seriously. (I understand that letters from China may not be written by the senders. I was asked to draft a letter by Prof. L when I asked him for a letter to send me to the US in 1985. This is also not uncommon for US professors.) If I know the professor who wrote the letter and respect him, and if the recommendation is strong, I would pick his student/staff. Nothing else matters. It's that simple. If I don't know the people who wrote the letter, I would ask my colleagues who may know the letter sender. Again, you can see how much I rely on recommendation. Too much, maybe. If an applicant is from a country that I don't have any knowledge, I would reach out to colleagues from this country. In fact, I was asked a lot about Chinese students during my first 10 years in the US, as if I was supposed to know every Chinese in physical oceanography. (I actually did know many of them.) GPAs should matter, but not as much as the letter. The ranking of the school where the student is/was? I don't think so; probably because I myself was not from a top-10 school. By the way, I don't like such email, starting with "Dear Sir, ..." If a student is writing to ME, he or she should at least know who I am. (I know I am nobody, but if you want to be my student, don't you think you should know what I do for a living???) To wrap up, and to make you feel you didn't really waste all your time reading this Blog, I did have funding for a post doc once, many years ago. How did I pick him out of N applications? The letter; yes, the letter from a well known colleague in the field.
An IncentiveCompatible Distributed Recommendation Model Jose M. Vidal Proceeding of to 6th International Workshop on Trust, Privacy, Deception, and Fraud in Agent Societies, 2003 ABSTRACT Our research is concerned with the study and development of incentive-compatible exchange mechanisms for recommendations in a multiagent system. These mechanism will allow and motivate agents to create an economy of ideas, where agents trade recommendations between themselves. In this paper we present a domain model and an incentivecompatible protocol for information exchange. Our model captures a subset of the realities of recommendation exchanges in the Internet. We provide an algorithm that selfish agents can use for deciding whether to exchange recommendations and with whom, that is, they can decide who they can trust to provide adequate recommendations. We analyze this algorithm and show that, under certain common circumstances, the agents’ rational choice is to exchange recommendations. Finally, we have implemented our model and algorithm and tested the performance of various populations. Our results show that both the social welfare and the individual utility of the agents is increased by participating in the exchange of recommendations. General Terms: Learning in Multiagent Systems, Recommender Systems, Trust 1. INTRODUCTION recommender systems the drawback of recommender systems: centralized implementations. the author's goal : to find ways to enable the emergence of a truly distributed recommender system in an Internet populated by selfish agents. incentive protocol design the research focus: under which circumstances should an agent give a recommendation to another? the organization of this paper: 1) Section2 describes a formal model to abstract the problem and its situation. 2) Section 3 derive some analytical results from this model. 3) Section 4 show the model simulation and test results. 4) Section 5 shows some work related to our research. 2. THE MODEL formal logic the utility i receives for reading d is given by: the expected value of Bayes Theorem: Tit-for-Tat to be the evolutionary stable strategy a Prisoner’s Dilemma matrix 2.1 An Agent’s Choice 3. MODELING USER PREFERENCES Netlogo 4. IMPLEMENTATIONANDTESTRESULTS source code: http://jmvidal.cse.sc.edu/netlogomas/ 4.1 Standard Agents clustering probability 4.2 Greedy Agents 4.3 Results Summary communities 5. RELATED WORK PHOAKS , the Referral Web 6. CONCLUSION author's work: 1) presented a domain model that captures the most important aspects of the distributed recommendations scenario\ 2) analyzed this model and showed that engaging in an exchange is the rational choice as long as the agent believes that the other agent has interests that have proven to be sufficiently similar 3) gave an algorithm that agents can use for deciding when to exchange recommendations and with whom 4) tested our algorithm on a various simulated scenarios. I comment: I don't understand the meanings of this paper distributedrec.nlogo An IncentiveCompatible.pdf Meanwhile, I read some code of 于同奎, but I don't comprehensive its essence. 惩罚与合作——一个来自人工社会的启示.pdf