# Web Science & Human Dynamics | Bruno Gonçalves

Source: https://bgoncalves.com/web-science/

> Bruno Gonçalves's research on Web traffic and human dynamics — why PageRank fails as a model of browsing, session structure, and agent-based models of navigation.

## Web Science & Human Dynamics

How people actually browse the Web, and why the PageRank random surfer gets it wrong.

## Overview

Between 2008 and 2010, working with complete HTTP logs from a university population, Bruno Gonçalves studied how people actually navigate the Web — not how link-graph models assume they do. The recurring finding across this work is that the random surfer at the heart of PageRank is a poor description of real browsing, and that reproducing genuine traffic requires agents with topical interests and memory.

## Research Areas

- **Beyond PageRank**

  Empirical Web traffic turns out to be a poor match for PageRank and other Markovian surfer models — real navigation carries structure those models cannot reproduce.

- **Session Structure**

  Recovering what a browsing session actually looks like from complete HTTP request logs, with user identity preserved across requests.

- **Agent-Based Models**

  Building generative models of browsing in which agents carry topical interests and memory of pages they liked, reproducing traffic patterns that random-surfer models miss.

- **Individual Dynamics**

  Characterizing single users through their return times to a page and the intervals between consecutive visits, rather than through aggregate statistics.

## PageRank Is Not How People Browse

Analysis of aggregate and individual Web requests showed that PageRank is a poor predictor of traffic. Using empirical patterns generated by a thousand users over two months, this work characterized properties of Web traffic that Markovian models cannot reproduce — in both aggregate statistics such as page traffic and in the structure of individual browsing. The proposed alternative gives each agent topical interests and a memory of pages it has liked, which recovers the observed behavior far better.

Preserving user identity across the request log — unusual for the period — is what made the individual-level results possible, including the distribution of return times to a page and the intervals between consecutive visits.

## Key Publications

- **[Modeling Traffic on the Web Graph](https://bgoncalves.com/pdf/meiss2010-1.pdf)**

  Meiss, **Gonçalves**, Ramasco, Flammini, Menczer. WAW / LNCS 6516 (2010)

- **[Agents, Bookmarks and Clicks: A topical model of Web traffic](https://arxiv.org/abs/1003.5327)**

  Meiss, **Gonçalves**, Ramasco, Flammini, Menczer. ACM Hypertext (2010)

- **[What’s in a Session: Tracking Individual Behavior on the Web](https://arxiv.org/abs/1003.5325)**

  Meiss, Duncan, **Gonçalves**, Ramasco, Menczer. ACM Hypertext (2009)

- **[Remembering what we like: Toward an agent-based model of Web traffic](https://arxiv.org/abs/0901.3839)**

  **Gonçalves**, Meiss, Ramasco, Flammini, Menczer. WSDM Late Breaking Results (2009)

- **[Towards the characterization of individual users through Web analytics](https://arxiv.org/abs/0901.0498)**

  **Gonçalves**, Ramasco. Complex Sciences (2009)

- **[Human Dynamics Revealed Through Web Analytics](https://arxiv.org/abs/0803.4018)**

  **Gonçalves**, Ramasco. Physical Review E (2008)

A complete list is available on the [Publications page](https://bgoncalves.com/publications).
