pdf AD HOC DIFFICULTY ESTIMATION FOR CROWDSOURCING SEMANTIC SIMILARITY Популярные

Усталов Дмитрий Алексеевич

Екатеринбург, Россия

The phenomenon of crowdsourcing is often called artificial artificial intelligencedue to the possibility of emulating the computation systems’ performance withhuman workers. Such a mechanistic approach is applied in Crowdforge that representshuman-computed MapReduce [1], by the CrowdDB extending the SQLsyntax with the keyword CROWD that submits the corresponding tasks to a crowdsourcingplatform [2], by the human-assisted data analysis system CDAS [3], etc.Nevertheless, online crowdsourcing is a quite complex system that still has manyunexploited opportunities [4].Since modern natural language processing methods require many statisticaldata to train a model, it is becoming much more difficult to process Russian,a resource-poor language [5]. Hence, it is reasonable to apply crowdsourcing forfacilitating the language resources.However, many researchers agree that one of the most challenging problemsin crowdsourcing is the cold start problem [6]. The cold start problem may beaddressed by estimating parameters using information from the task domain.The present study is dedicated to this problem and is focused on a quite simplehuman computation task—semantic similarity assessment.