Using first- and second-order lexical co-occurrence to assess temporal changes in media valence and content

加拿大人受益于自然资源产生的社会、环境和经济可持续性， 资源加拿大自然资源潜力的全部价值只有在政府和 工业界可以积极地共同努力，实现对未来的共同愿景， 利益攸关方往往对自然资源开发的最佳方法持有非常不同的意见。 这项建议的目的是制定一种方法，使工业界更容易监测和了解 利益相关者对自然资源开发的不同观点。我建议使用自动化 （计算）方法从媒体文本中提取内容，以开发定量方法， 公众对加拿大资源的态度。特别是，我将使用认知方法， 一门通过测量微观语境的相似性来模拟单词语义内容的科学， 这些话出现了。这些模型为语言中的每个单词构建向量（只是数字列表 其表示该单词与小文本窗口内每隔一个单词共同出现频率，对于 例如，在任一方向上包括五个单词。通过数学比较向量的相似性， 任何两个单词，都可以测量任何两个单词的相似程度。通过结合一个词的相似性， 两个或多个单词，可以根据预定义的语义维度来表征单词。为 例如，在以前的工作中，我已经表明，人们可以查看一个词与具有高 情感效价（即被认为是非常好或非常坏的词），以根据其 评价效价我将使用语义共现模型和其他计算的文本度量， 描述不同的印刷媒体和对自然资源有既得利益的利益相关者群体 改变他们在不同时间交流这些资源的方式

Canadians benefit from the social, environmental, and economic sustainability generated from natural resources. The full value of Canada's natural resource potential can best be realized when government and industry can actively work together toward a shared vision of the future that takes into account the views of stakeholders who often hold very different opinions about the best approach to natural resource development. The purpose of this proposal is to develop a method to make it easier for industry to monitor and understand the differing points of view of the stakeholders' in natural resource development. I propose to use automated (computational) approaches to extract content from media text to develop quantitative methods of characterizing public attitudes about Canadian resources. In particular, I will use methods from cognitive science which model semantic content of words by measuring the similarity of the micro-contexts in which those words appear. These models build vectors (which are just lists of numbers) for each word in the language that represent how often that word co-occurs with every other word within a small text window, that may, for example, include five words in either direction. By mathematically comparing the similarity of the vectors for any two words, it is possible to measure how similar any two words are. By combining a word's similarities to two or more words, it is possible to characterize words in terms of predefined semantic dimensions. For example, in previous work I have shown that one can look at a word's relative distance from words with a high emotional valence (i.e. words considered to be very good or very bad) to characterize that word in terms of its evaluative valence. I will use semantic co-occurrence models and other computed measures of text to characterize how different print media outlets and stakeholder groups with a vested interest in natural resources change the way they communicate about those resources across time