CiteULike: pdlug Zhou

Ultra accurate personal recommendation via eliminating redundant correlations

Tao Zhou — 2008-05-28T14:54:51-00:00

(27 May 2008)

In this paper, based on a weighted projection of bipartite user-object network, we introduce a personal recommendation algorithm which has remarkably higher accuracy than the classical algorithm, namely collaborative filtering. In this algorithm, the correlation resulting from a specific attribute may be repeatedly counted in the cumulative recommendations from different objects. By considering the higher order correlations, we design an effective algorithm that can, to some extent, eliminate the redundant correlations. The algorithmic accuracy, measured by the ranking score, can be further improved by 23% in the optimal case. Most of the previous studies considered the algorithmic accuracy only, in this paper, we argue that the diversity and popularity, as two significant criteria of algorithmic performance, should also be taken into account. With more or less the same accuracy, an algorithm giving higher diversity and lower popularity is more favorable. Numerical results show that the present algorithm can outperform the standard one simultaneously in all three criteria: higher accuracy, higher diversity, and lower popularity.

On the probability distribution of stock returns in the Mike-Farmer model

Gao-Feng Gu — 2008-05-27T02:34:16-00:00

(23 May 2008)

Recently, Mike and Farmer have constructed a very powerful and realistic behavioral model to mimick the dynamic process of stock price formation based on the empirical regularities of order placement and cancelation in a purely order-driven market, which can successfully reproduce the whole distribution of returns, not only the well-known power-law tails, together with several other important stylized facts. There are three key ingredients in the Mike-Farmer (MF) model: the long memory of order signs characterized by the Hurst index $H_s$, the distribution of relative order prices $x$ in reference to the same best price described by a Student distribution (or Tsallis' $q$-Gaussian), and the dynamics of order cancelation. They showed that different values of the Hurst index $H_s$ and the freedom degree $α_x$ of the Student distribution can always produce power-law tails in the return distribution $f(r)$ with different tail exponent $α_r$. In this paper, we study the origin of the power-law tails of the return distribution $f(r)$ in the MF model, based on extensive simulations with different combinations of the left part $f_L(x)$ for $x<0$ and the right part $f_R(x)$ for $x>0$ of $f(x)$. We find that power-law tails appear only when $f_L(x)$ has a power-law tail, no matter $f_R(x)$ has a power-law tail or not. In addition, we find that the distributions of returns in the MF model at different timescales can be well modeled by the Student distributions, whose tail exponents are close to the well-known cubic law and increase with the timescale.

Learning Multiple Graphs for Document Recommendations

Ding Zhou — 2008-05-09T04:14:29-00:00

(21 April 2008)

The Web offers rich relational data with different semantics. In this paper, we address the problem of document recommendation in a digital library, where the documents in question are networked by citations and are associated with other entities by various relations. Due to the sparsity of a single graph and noise in graph construction, we propose a new method for combining multiple graphs to measure document similarities, where different factorization strategies are used based on the nature of different graphs. In particular, the new method seeks a single low-dimensional embedding of documents that captures their relative similarities in a latent space. Based on the obtained embedding, a new recommendation framework is developed using semi-supervised learning on graphs. In addition, we address the scalability issue and propose an incremental algorithm. The new incremental method significantly improves the efficiency by calculating the embedding for new incoming documents only. The new batch and incremental methods are evaluated on two real world datasets prepared from CiteSeer. Experiments demonstrate significant quality improvement for our batch method and significant efficiency improvement with tolerable quality loss for our incremental method.

Multifractality in stock indexes: Fact or fiction?

Zhi-Qiang Jiang — 2007-06-15T14:56:32-00:00

(14 Jun 2007)

Multifractal analysis and extensive statistical tests are performed upon intraday minutely data within individual trading days for four stock market indexes (including HSI, SZSC, S&P500, and NASDAQ) to check whether the indexes (instead of the returns) possess multifractality. We find that the mass exponent $τ(q)$ is linear and the singularity $α(q)$ is close to 1 for all trading days and all indexes. Furthermore, we find strong evidence showing that the scaling behaviors of the original data sets cannot be distinguished from those of the shuffled time series. Hence, the so-called multifractality in the intraday stock market indexes is merely an illusion.