<div dir="ltr"><span style="font-size:12.8px">Hi everyone!</span><div style="font-size:12.8px"><br></div><div style=""><p class="MsoNormal" style="font-size:12.8px;background-image:initial;background-position:initial;background-size:initial;background-repeat:initial;background-origin:initial;background-clip:initial"><span style="font-size:10pt;line-height:15.3333px;font-family:Arial,sans-serif">This week at the<span class="gmail-m_-2314642548440317959gmail-m_-5235992900627905237gmail-m_-5397734049088859634gmail-m_-8409739297008502631gmail-m_-5223976695426397292gmail-m_-8239203266232787201gmail-m_-3131546366916696156gmail-apple-converted-space"> </span></span>Applied Statistics <span style="line-height:14.72px"><font face="Arial, sans-serif"><span style="font-size:10pt">Worksho<wbr>p we will be welcoming</span></font><span class="gmail-m_-2314642548440317959gmail-m_-5235992900627905237gmail-m_-5397734049088859634gmail-m_-8409739297008502631gmail-m_-5223976695426397292gmail-m_-8239203266232787201gmail-m_-3131546366916696156gmail-apple-converted-space" style="font-family:Arial,sans-serif;font-size:10pt"> <b><i>Christopher Lucas</i></b></span><font face="Arial, sans-serif"><span style="font-size:10pt">, graduate student in the Harvard Government Department. He will be presenting his job talk entitled</span></font><b style="font-family:Arial,sans-serif;font-size:10pt"><i> </i></b><font face="Arial, sans-serif"><span style="font-size:13.3333px"><b><i>A Model for Political Video: The Audio Video Neural Network</i></b></span><span style="font-size:10pt">.  Please find the abstract below and on the Applied Stats website <a href="https://projects.iq.harvard.edu/applied.stats.workshop-gov3009" target="_blank">here</a>.</span></font></span></p><p class="MsoNormal" style="font-size:12.8px;background-image:initial;background-position:initial;background-size:initial;background-repeat:initial;background-origin:initial;background-clip:initial"><span style="font-size:10pt;line-height:15.3333px;font-family:Arial,sans-serif"> </span></p><p class="MsoNormal" style="font-size:12.8px;background-image:initial;background-position:initial;background-size:initial;background-repeat:initial;background-origin:initial;background-clip:initial"><span style="font-size:10pt;line-height:15.3333px;font-family:Arial,sans-serif">As usual, we will meet at noon in CGIS Knafel Room 354 and lunch will be provided.  See you all there!<br clear="all"></span></p><div style="font-size:12.8px"><div class="gmail_signature"><div dir="ltr"><br><div>-- Dana Higgins</div></div></div></div>
<p style="font-size:12.8px"></p><p class="MsoNormal" style="font-size:12.8px;background-image:initial;background-position:initial;background-size:initial;background-repeat:initial;background-origin:initial;background-clip:initial"><span style="font-size:10pt;line-height:15.3333px;font-family:Arial,sans-serif"> </span></p><p class="MsoNormal" style="font-size:12.8px"><br></p><p class="MsoNormal" style="font-size:12.8px"><span style="line-height:15.3333px"><b style="font-family:Arial,sans-serif;font-size:10pt"><span style="background-image:initial;background-position:initial;background-size:initial;background-repeat:initial;background-origin:initial;background-clip:initial">Title:</span></b><span class="gmail-m_-2314642548440317959gmail-m_-5235992900627905237gmail-m_-5397734049088859634gmail-m_-8409739297008502631gmail-m_-5223976695426397292gmail-m_-8239203266232787201gmail-m_-3131546366916696156gmail-apple-converted-space" style="font-family:Arial,sans-serif;font-size:10pt"><span style="background-image:initial;background-position:initial;background-size:initial;background-repeat:initial;background-origin:initial;background-clip:initial"> </span></span><u style="font-family:Arial,sans-serif;font-size:10pt"><span style="background-image:initial;background-position:initial;background-size:initial;background-repeat:initial;background-origin:initial;background-clip:initial">A Model for Political Video: The Audio Video Neural Network</span></u><br><br></span></p><p class="MsoNormal" style=""><span style="font-size:12.8px;line-height:15.3333px"><br><b style="font-family:Arial,sans-serif;font-size:10pt"><span style="background-image:initial;background-position:initial;background-size:initial;background-repeat:initial;background-origin:initial;background-clip:initial">Abstract:</span></b><span class="gmail-m_-2314642548440317959gmail-m_-5235992900627905237gmail-m_-5397734049088859634gmail-m_-8409739297008502631gmail-m_-5223976695426397292gmail-m_-8239203266232787201gmail-m_-3131546366916696156gmail-apple-converted-space" style="font-family:Arial,sans-serif;font-size:10pt"><span style="background-image:initial;background-position:initial;background-size:initial;background-repeat:initial;background-origin:initial;background-clip:initial"> </span></span></span><span style="font-size:12.8px"> </span>In the 2016 election cycle political campaigns spent over $4.4 billion on television advertisements. What type of content do campaigns use in their efforts to influence voter behavior? Past efforts to answer questions such as these, where media and television content are the topic of interest, require researchers to engage in herculean efforts to manually watch and classify video content. This is time consuming and prohibitively expensive for most researchers. I develop a novel, general approach to video classification, the Audio-Video Neural Network (AVNN), which is the first contribution from political science to deep learning. The AVNN recovers subtle categories of interest to political scientists like “fear” and “negativity” while also successfully learning to parse more topical classes like “political advertisement” and “cable news.” Importantly, the model I propose can learn from both visual and audio features. I demonstrate the AVNN by analyzing campaign advertisements in American elections, first showing that my model can be used to separate political campaign advertisements from cable news. Second, I show how subtle video labels like “appeals to fear” can also be recovered. All methods described here are implemented in easy-to-use Rand Python packages</p></div>
</div>