Timeline for How good is my recommender? [duplicate]
Current License: CC BY-SA 4.0
12 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jun 25, 2019 at 14:36 | history | closed |
mkt mdewey kjetil b halvorsen♦ Sycorax♦ Siong Thye Goh |
Duplicate of Is my model any good, based on the diagnostic metric ($R^2$/ AUC/ accuracy/ RMSE etc.) value? | |
| Jun 24, 2019 at 11:40 | review | Close votes | |||
| Jun 25, 2019 at 14:36 | |||||
| Jun 18, 2018 at 17:46 | answer | added | Tim | timeline score: 2 | |
| Jun 18, 2018 at 16:32 | comment | added | Sextus Empiricus | I think that a question about 'how should my recommender system perform' is though or impossible to answer (possibly you could compare your algorithm with some others but it remains a specific case). You could however ask about 1) Extrinsic, what are some of the more famous/common recommender algorithms and applications of them and how do they perform? (where coomon/famous remains a bit ambiguous) or 2) Intrinsic, what type of methods can I use to express the value of my recommender system. ( not caring about other potential better ones but just wanting to express that the current is good) | |
| Jun 18, 2018 at 16:12 | answer | added | Kodiologist | timeline score: 3 | |
| Jun 18, 2018 at 16:06 | comment | added | PolBM | I think is allways good to try with some benchamrks. Such as 1) reomend allways the most popular item, 2) recomend the last viewed item of that user. This way you can check which precision those benchmarks have and by how much your recomender is imporving it. | |
| Jun 18, 2018 at 15:54 | comment | added | Sycorax♦ | This is an article about Netflix's recommender system. dl.acm.org/citation.cfm?id=2843948 Looks promising. | |
| Jun 18, 2018 at 15:52 | comment | added | User12547645 | @Sycorax So far I did not find recent literature, which evaluates the accuracy of recommender systems. All I am interested in is how does the accuracy of this model compare to e.g. Amazon, Netflix or anyone else. I really have no idea, if 20% are awesome or terrible | |
| Jun 18, 2018 at 15:48 | comment | added | Sycorax♦ | This doesn't really seem answerable in a context-free setting. Is your model good enough for what purpose? You could conduct a literature review and find publications claiming to obtain some level of precision using some model & compare your results. Or, if you're trying to make money using recommendations, then you can set a certain target for sales. Or, if you're doing this for a class, you need to make sure this project satisfies the rubric. I don't think anyone can answer these questions for you. | |
| Jun 18, 2018 at 15:46 | comment | added | User12547645 | @Sycorax Thank you very much for the hin. You are right, the language and framework are not really important here. Do you have an answer to the question, though | |
| Jun 18, 2018 at 15:40 | history | edited | Sycorax♦ |
edited tags
|
|
| Jun 18, 2018 at 15:38 | history | asked | User12547645 | CC BY-SA 4.0 |