Collaborative Topic Ranking: Leveraging Item Meta-Data for Sparsity Reduction
Authors: Weilong Yao, Jing He, Hua Wang, Yanchun Zhang, Jie Cao
AAAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct an experimental study on a real and publicly available dataset, and the results show that our algorithm is effective in providing accurate recommendation and interpreting user factors and item factors. and We conduct an experimental study on a large and real-world dataset. The results empirically show that CTRank outperforms other recommendation methods, in terms of prediction accuracy. |
| Researcher Affiliation | Academia | 1University of Chinese Academy of Sciences, Beijing, China 2Centre for Applied Informatics, College of Engineering & Science, Victoria University, Australia 3Fudan University, Shanghai, China 4 Nanjing University of Finance and Economics, Nanjing, China |
| Pseudocode | No | The paper describes the parameter learning process in narrative text with mathematical equations, but it does not include a distinct block, figure, or section explicitly labeled as "Pseudocode" or "Algorithm". |
| Open Source Code | No | The paper does not provide any specific link or explicit statement indicating that the source code for the proposed CTRank method is openly available. It only mentions the dataset Cite ULike is public. |
| Open Datasets | Yes | Dataset Cite ULike1 is an academic social network, which allows users to create individual reference libraries for the articles they like. In this work, we use a large dataset collected by Wang (Wang and Blei 2011) from Cite ULike. Articles in a user s reference library are considered as observed items. In this dataset, 5551 users expressed 204,986 observed ratings for 16,980 items (articles) with a high sparseness of 99.78%. ... 1http://www.citeulike.org |
| Dataset Splits | Yes | Following one-class CF literature (Rendle et al. 2009; Hariri, Mobasher, and Burke 2012; Paquet and Koenigstein 2013), we adopt the popular 10-fold leave-one-out cross validation to evaluate the performance of recommendation models. That is, we randomly remove one example from Cite ULike dataset R+ to form test set Rtest, and the remaining constitutes the training set Rtrain. The models are learned on Rtrain and their predicted ranking is evaluated on Rtest by averaging Hit Ratio@N, the probability that the removed article is recommended as part of the top-N recommendations: ... We repeat this process 10 times and report the average results. and Specifically, we randomly select one test example per user from training set and use these examples as validation set to determine the trade-off parameters {K, λv, λu}. |
| Hardware Specification | No | The paper does not provide any specific hardware details such as GPU models, CPU types, or cloud instance specifications used for running its experiments. |
| Software Dependencies | No | The paper does not provide specific software dependency details with version numbers (e.g., library or solver names with versions). |
| Experiment Setup | Yes | Parameter Settings For all comparison methods, we set respective optimal parameters either according to corresponding references or based on our validation experiments. Specifically, we randomly select one test example per user from training set and use these examples as validation set to determine the trade-off parameters {K, λv, λu}. For all these latent factor based methods, we set the dimensionality of latent space K = 200 and the learning rate η = 0.005. For BPR, parameters λu and λv are 0.0025. For PALS λv = λu = 0.0025. For CTR, λu and λv are set to 0.01 and 100, respectively. For CTRank models, λv = 0.025, λu = 0.0025. |