基于话题的品牌形象认知及情感分析

发布时间：2019-06-20 22:14

【摘要】：品牌形象挖掘是了解品牌形象的关键步骤,是塑造提升品牌形象、制定品牌传播策略的第一步,对品牌形象构建以及品牌竞争都具有深远的意义。随着互联网技术的高速发展,用户随时随地在接触和获取信息,并创建了大量的用户生成内容。品牌传播被置身于更加灵敏便捷、自由高速却又充满不确定性的传播空间中。传统的品牌形象调查,在样本的多样性和时效性、以及分析方法上已不能满足品牌形象挖掘的需求。在这种环境下,海量碎片化的用户生成话题数据为品牌形象挖掘提供了丰富可行的数据资源和新的研究思路。基于用户生成的品牌相关话题挖掘品牌形象,理解品牌在用户心中形成的认知和情感共鸣,是新传播环境下品牌形象传播的战略基础。用户形成的品牌形象包括用户对品牌的认知、情感、行为三个方面,本文只关注认知和情感两个维度。首先给出面向品牌形象挖掘的话题识别方法,从而针对不同的分析诉求,对所需的数据范围进行取舍。其次,设计面向品牌形象认知和情感的挖掘方法,从海量、碎片化的用户生成话题数据中获取用户对品牌形象的认知和用户对品牌形象的个体情感以及群体情感状态。论文的具体研究内容如下:(1)面向品牌形象挖掘的话题识别方法。品牌形象挖掘使用的数据范围因任务侧重点不同而不同。论文面向品牌形象挖掘给出不同的话题识别方法,首先基于关键词搜索品牌相关的常规话题,给出品牌关注度时序曲线,企业可以根据关注度差异选取数据范围。其次,基于曲线分类建模的思想预测热门话题,适应高时效性要求的品牌形象挖掘任务。热门话题发现的基本思想是从话题的统计特性出发,使用传播扩散程度和关注聚焦程度刻画话题的热度,建立话题的热度曲线。通过对话题的热度曲线进行预处理,消除原始量纲对热度曲线内在相似性判定带来的负面影响。并对丰富多变的曲线进行分类建模,从中提取共性特征和行为规律,使之呈现出较为明朗的规律性。应用热度曲线分类模型上的加权投票规则预测新话题是否会发展成热门话题。基于关键词搜索常规话题和基于曲线分类建模识别热门话题可以满足品牌形象挖掘对数据选择的一般要求。(2)基于话题的品牌形象认知分析方法。用户对品牌形象的认知指的是用户对品牌的整体印象(包括功能、服务、效用等的评价),是品牌形象传播的基础。论文提出一种基于规则的认知标签提取方法,从用户生成内容中掌握用户对品牌形象的认知。首先,基于语言规则提取出初始的认知标签;然后,借助于同义词词典和Jaccard相似度对认知标签进行聚合;最后,应用TFMF模型计算聚合后不同认知标签的重要性。根据所获取的重要认知标签,企业能够更好的理解消费者对品牌的整体印象、最在意的品牌特性以及与竞品相比品牌所拥有的独特属性。(3)基于话题的品牌形象个体情感分析方法。古语有云:“攻心为上”。情感是品牌传播的攻心武器,对品牌形象挖掘离不开对用户个体情感状态的把握。有效提取用户生成内容的情感标签是品牌形象个体情感分析的基础。新词的涌现、热词的漂移、海量碎片化及中文常用词特性带来的高维稀疏性成为中文情感分类的主要困难。论文提出一种新颖的方法用以解决上述问题:构造表情符号词典用来自动获取训练集情感标签,解决海量数据的标注问题。这样可以节省训练标签所需的人力和财力成本,且具有较高的客观性。引入修正的G2检验联合情感词词典进行特征选择,该方法可以保留强分类能力的特征而不至于过过滤,并尽可能消除无效特征的干扰,从而进行降维,控制稀疏性。采用多阶段判断式的抽样策略生成训练集,保证基分类器的多样性。最后采用加权多数投票的方式对基分类器结果进行融合,解决特征和情感漂移及碎片化问题。实验表明该方法可以快速有效的获取训练标签,保留下强区分能力的特征,并实现较高的精度。且该方法很容易扩展到流数据并实现并行化。(4)基于话题的品牌形象群体情感分析方法。情感作为消费体验中最重要的角色,理解用户对品牌的群体情感状态以及群体情感演化逻辑可以帮助企业和用户理解品牌形象。本章构造品牌群体情感计量模型,基于个体情感对群体情感进行集结。建立不同粒度下群体情感时间序列数据,通过对群体情感时序数据的分析,理解品牌群体情感演化的逻辑。分析热门话题的属性,了解热门话题对品牌群体情感演化的影响。通过实验以及案例分析可以发现,品牌生命周期的不同阶段会带来品牌群体情感的不同状态。热门话题会加速情感演化的过程,且热门话题的不同属性会影响群体情感演化的方向,而企业处理策略会加深或者消解热门话题带来的情感演化影响程度。品牌形象挖掘是品牌形象传播的重要基础。论文从品牌形象挖掘的数据准备,品牌形象认知的挖掘方法、品牌形象情感的挖掘方法三个方面开展研究,帮助企业从海量碎片化的数据中提取品牌形象,理解用户对品牌形象的感知,进而构建品牌的核心竞争力。
[Abstract]:Brand image mining is a key step in understanding the brand image. It is the first step to build up the brand image and develop the brand communication strategy. It is of far-reaching significance to the brand image construction and the brand competition. With the high-speed development of the Internet technology, users contact and obtain information at any time and any place, and a large number of user-generated content is created. The spread of the brand is in a more sensitive, convenient, free, high-speed and uncertain propagation space. The traditional brand image survey, in the diversity and timeliness of the samples, and the analysis method cannot meet the requirement of brand image mining. In this environment, the user-generated topic data of massive defragmentation provides rich and feasible data resources and new research ideas for brand image mining. Based on the user-generated brand-related topic mining brand image, the understanding of the brand's cognitive and emotional resonance in the user's heart is the strategic foundation of the brand image transmission in the new communication environment. The brand image formed by the user includes three aspects of the user's cognition, emotion and behavior of the brand, and the article only concerns the two dimensions of cognition and emotion. Firstly, the topic identification method for brand image mining is given, so that the required data range is selected for different analysis demands. Secondly, the design of the method for the recognition and the emotion of the brand image is designed, and the user's perception of the brand image and the individual emotion and the group emotional state of the user on the brand image are obtained from the mass and fragmented user-generated topic data. The specific content of the thesis is as follows: (1) The topic identification method for brand image mining. The data range used for brand image mining is different from the task focus. In this paper, different topic identification methods are given for brand image mining. First, based on the general topic related to the keyword search brand, the time series curve of the brand attention is given, and the enterprise can select the data range according to the difference of the degree of attention. Secondly, the concept of model based on curve classification is a hot topic, which is suitable for the task of brand image mining with high timeliness. The basic idea of the hot topic discovery is to set up the heat curve of the topic based on the statistical character of the topic, using the degree of spread and the degree of focus and the degree of focus. By pre-processing the heat curve of the topic, the negative effect of the original dimension on the similarity determination of the heat curve is eliminated. And the rich and changeable curve is classified and modeled, and the common characteristic and the behavior rule are extracted from the curve, so that the characteristic is more clear. It is a hot topic to predict whether a new topic can be developed by using the weighted voting rule on the heat curve classification model. The general requirements of brand image mining on data selection can be met based on key word search routine and curve-based classification modeling. (2) The cognitive analysis method of brand image based on the topic. The user's perception of the brand image refers to the user's overall impression of the brand (including the evaluation of function, service, utility, etc.), which is the basis for the transmission of the brand image. In this paper, a rule-based method for extracting a cognitive tag is proposed, and the user's perception of the brand image is grasped from the user-generated content. First, the initial cognitive tag is extracted based on the language rule; then, the cognitive tag is aggregated with the aid of the synonym dictionary and the Jaccard similarity; and finally, the importance of different cognitive labels after the aggregation is calculated by using the TFMF model. According to the important cognitive label acquired, the enterprise can better understand the overall impression of the brand by the consumer, the brand characteristic which is the most important, and the unique property owned by the brand as compared with the competitor. (3) The individual emotion analysis method of the brand image based on the topic. The ancient language has the cloud: the "to attack one's heart". The emotion is the core weapon of the communication of the brand, and it is necessary to grasp the individual emotion state of the user without the brand image mining. The emotional label that effectively extracts the user-generated content is the basis of the individual emotion analysis of the brand image. The emergence of new words, the drift of hot words, the massive defragmentation and the high-dimensional sparsity brought by the characteristics of Chinese language often become the main difficulty of Chinese sentiment classification. In this paper, a novel method is proposed to solve the above problems: the structure of the emoticon dictionary is used to automatically acquire the emotional label of the training set and solve the problem of the labeling of the mass data. So that the labor and financial cost required by the training label can be saved, and the method has higher objectivity. A modified G2-test combined affective word dictionary is introduced for feature selection. The method can retain the characteristics of strong classification capability without filtering, and eliminate the interference of the invalid features as much as possible, thereby reducing the dimension and controlling the sparsity. And a multi-stage judgment-type sampling strategy is adopted to generate a training set, so that the diversity of the base classifier is guaranteed. And finally, the basis classifier results are fused in a weighted majority vote mode to solve the problems of characteristic and emotion drift and fragmentation. The experiment shows that the method can quickly and effectively obtain the training label, keep the characteristic of the strong distinguishing ability, and realize the higher precision. And the method is easy to extend to the stream data and realize the parallelization. (4) The emotion analysis method of the brand image group based on the topic. Affective as the most important role in the consumption experience, it is to be understood that the user's emotional state of the group and the logic of the group's emotional evolution can help enterprises and users to understand the brand image. This chapter constructs the emotional measurement model of the brand group, and builds the group's emotion based on the individual emotion. The data of the group's emotional time series under different granularities is established, and the logic of the emotional evolution of the brand group is understood through the analysis of the group's emotional time series data. The paper analyzes the properties of the hot topic and the influence of the hot topic on the emotional evolution of the brand group. Through the experiment and case analysis, it can be found that the different phases of the brand life cycle can bring different states of the brand group's emotion. The hot topic can accelerate the process of the emotional evolution, and the different attributes of the hot topics affect the direction of the group's emotional evolution, and the enterprise's processing strategy will deepen or eliminate the influence of the emotional evolution brought by the hot topic. Brand image mining is an important basis for the transmission of brand image. From the data preparation of brand image mining, the mining method of brand image cognition and the mining method of brand image emotion, this paper helps enterprises to extract the brand image from the data of massive defragmentation, and to understand the user's perception of the brand image. And then building the core competitiveness of the brand.
【学位授予单位】：合肥工业大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：F273.2

【参考文献】