Yaniv Dover

Promotional Reviews: An Empirical Investigation of Online Review Manipulation

Online reviews could, in principle, greatly improve the match between consumers and products. However, the authenticity of online user reviews remains a concern; firms have an incentive to manufacture positive reviews for their own products and negative reviews for their rivals. In this paper, we marry the diverse literature on economic subterfuge with the literature on organizational form. We undertake an empirical analysis of promotional reviews, examining both the extent to which fakery occurs and the market conditions that encourage or discourage promotional reviewing activity. Specifically, we examine hotel reviews, exploiting the organizational differences between two travel websites: Expedia.com, and Tripadvisor.com. While anyone can post a review on Tripadvisor, a consumer could only post a review of a hotel on Expedia if the consumer actually booked at least one night at the hotel through the website. We examine differences in the distribution of reviews for a given hotel between Tripadvisor and Expedia. We show in a simple model that the net gains from promotional reviewing are likely to be highest for independent hotels that are owned by single-unit owners and lowest for branded chain hotels that are owned by multi-unit owners. Our methodology thus isolates hotels with a disproportionate incentive to engage in promotional reviewing activity. We show that hotels with a high incentive to fake have a greater share of five star (positive) reviews on Tripadvisor relative to Expedia. Furthermore, we show that the hotel neighbors of hotels with a high incentive to fake have more one and two star (negative) reviews on Tripadvisor relative to Expedia.

Network Traces on Penetration: Uncovering Degree Distribution from Adoption Data

We show how networks modify the diffusion curve by affecting its symmetry. We demonstrate that a networks degree distribution has a significant impact on the contagion properties of the subsequent adoption process, and we propose a method for uncovering the degree distribution of the adopter network underlying the dissemination process, based exclusively on limited early-stage penetration data. In this paper we propose and empirically validate a unified network-based growth model that links network structure and penetration patterns. Specifically, using external sources of information, we confirm that each network degree distribution identified by the model matches the actual social network that is underlying the dissemination process. We also show empirically that the same method can be used to forecast adoption using an estimation of the degree distribution and the diffusion parameters at an early stage (15%) of the penetration process. We confirm that these forecasts are significantly superior to those of three benchmark models of diffusion. Our empirical analysis indicates that under heavily right-skewed degree distribution conditions (such as scale-free networks), the majority of adopters (in some cases, up to 75%) join the process after the sales peak. This strong asymmetry is a result of the unique interaction between the dissemination process and the degree distribution of its underlying network.

A short account of a connection of power laws to the information entropy

We use the formalism of “maximum principle of Shannons entropy” to derive the general power law distribution function, using what seems to be a reasonable physical assumption, namely, the demand of a constant mean “internal order” (Boltzmann entropy) of a complex, self-interacting, self-organized system.

Do all economies grow equally fast

The stochastic spatially extended generalized Lokta-Volterra approach introduced in Solomon [48], Challet et al. [18], Yaari et al. [58], is extended to the study of interactions between economic sectors, countries and blocks. The theory predicts robustly in a very wide range of conditions systematic regularities in the growth rates evolution of various subsystems. The J-curve phenomenon which was studied in Challet et al. [18] is revisited and more empirical support is given to the theory. In particular to the connection between the economic minimum and the crossover of the new emergent leading sector with the old decaying one. We describe the ’Growth Alignment Effect’ (GAE), it’s theoretical basis and demonstrate it empirically for numerous cases in the inter-national and intra-national economies. The GAE is the concept that in steady state the growth rates of the GDP per capita of the various system components align. We differentiate the GAE predictions from the usual convergence or divergence conceptual framework. Further investigations of GAE and subsidiaries are suggested and possible uses are proposed. Due to it’s simple and robust nature, the method can be used as a tool for economic decisions and policy making.

Uncovering Social Network Structures through Penetration Data

We show how networks modify the diffusion curve by affecting its symmetry. We demonstrate that a networks degree distribution has significant impact on the contagion properties of the subsequent adoption process, and propose a method for uncovering the degree distribution of the adopter network underlying the dissemination process, based exclusively on limited early-stage penetration data. In this paper we propose and empirically validate a unified network-based growth model that links network structure and penetration patterns. Specifically, using external sources of information, we confirm that each network degree distribution identified by the model matches the actual social network that is underlying the dissemination process. We also show empirically that the same method can be used to forecast adoption using an estimation of the degree distribution and the diffusion parameters, at an early stage (15%) of the penetration process. We confirm that these forecasts are significantly superior to those of three benchmark models of diffusion. Our empirical analysis indicates that under heavily right-skewed degree distribution conditions (such as scale-free networks), the majority of adopters (in some cases, up to 75%) join the process after the sales peak. This strong asymmetry is a result of the unique interaction between the dissemination process and the degree distribution of its underlying network.

Nucleation of Social Groups: The Role of Centrality Inequality and Social Mobility

Even though the heterogeneity of social networks centrality is well documented, its role and effect on network stability, is unclear. It is known that, universally, networks have an “inner” highly-connected nucleus and, in contrast, sparser outer shells. But, to what extent the existence of this nucleus is crucial for the survival of a network? To what extent is the outer shells, much-larger population, essential to the longevity of the network? Furthermore, network structure is very much dependent on the mobility between centrality shells, i.e., social mobility. What is, then, the role of social mobility in the formation of the nucleus-periphery profile and does it have an effect on network lifetime? Here, we explore these questions using data collected of more than 10K networked communities, with more than 134K users, for over a decade. We find that: (i) social mobility is, on average, negative and promotes instability, and (ii) the more positive social mobility is, the more stable the community. Further, (iii) the network is composed of two phases, a large but ephemeral sparsely-connected cloud of actors which nucleates around a highly stable nucleus of core users. Finally, (iv) networked communities which closely maintain a nucleation ratio, i.e., ratio between nucleus size and outer shells size, of 1 to 3, exhibit the best chances of survival. Deviations from this nucleation ratio translates into the collapse of the network, especially for younger communities.

Pump It Out! The Effect of Transmitter Activity on Content Propagation in Social Media

People share billions of pieces of content such as news, videos, and photos through social media every day. Marketers are interested in the extent to which such content propagates and, importantly, which factors make widespread propagation more likely. Extant research considers various factors, such as content attributes (e.g., newness), source traits (e.g., expertise), and network structure (e.g., connectivity). This research builds on prior work by introducing a novel behavior-focused transmitter characteristic that is positively associated with content propagation in social media: activity, or how frequently a person transmits content. Evidence for this effect comes from five studies and different paradigms. First, two studies using data from large social media platforms (Twitter and LiveJournal) show that content posted by higher-activity transmitters — whom we refer to as “social pumps” — propagates more than content posted by lower-activity transmitters. Second, three experiments explore the mechanism driving this effect, showing that social media users receiving content from a social pump are more likely to retransmit it (a necessary behavior for achieving aggregate-level propagation) because they infer that content from a social pump is more likely to be current, and therefore more attractive as something to pass along through retransmission.

Constrained Simulations of the Local Universe

This project is part of an international collaboration between different institutions around the world with the aim of generating the most accurate simulations of the formation of the Local Universe from cosmological initial conditions. In these kind of simulations, the smaller Fourier modes of a random realization of the power spectrum of density fluctuations are substituted by observational constrains in order to recover our real environment. These observational data come from MARK III [1], SBF [2], nearby X-ray clusters of galaxies [3] and the Karachentsev catalogues [4]. Different N-body simulations, with different resolutions, box sizes, codes and goals, have been performed up to now. Our higest resolution run consists of a resimulated region of 2 h−1 Mpc radius around the Local Group candidate found in a computational box of Lbox = 64 h−1 Mpc for the ΛCDM model with WMAP3 parameters. Initial conditions were generated with 4096 particles. The particle mass in the high resolution area correspond to Mp = 2.54×105h−1 M . It lets us make an unprecedented detailed analysis of the dynamics and substructures of the Local Group members.