分析:GPT+SBERT做用研统计

分析:GPT+SBERT做用研统计

分析:GPT+SBERT做用研统计 分析:GPT+SBERT做用研统计 Modified January 11, 2024 2. 算法管线 def analyze sentiment(text): """ 对文本进行情感分析,长文本将被分割处理。 """ if len(text) <= 512: 假设512是模型的最大长度 return sentiment analyzer(text)[0]['label'] else: chunks = chunk text(text, 512) sentiments = [sentiment analyzer(chunk)[0]['label'] for chunk in chunks] return max(set(sentiments), key=sentiments.count) 返回最常见的情感 def calculate similarity(text, sentences): """ 计算文本与给定句子之间的相似度。 """ case embedding = sbert model.encode(text, convert to tensor=True) sentences embeddings = sbert model.encode(sentences, convert to tensor=True) similarities = util.pytorch cos sim(case embedding, sentences embeddings)[0] return similarities.cpu().numpy() 对每个案例执行情感分析 df['Sentiment'] = df['Case'].apply(analyze sentiment) 对每个案例计算与给定句子的相似度,并将句子作为列标题 for sentence in sentences to compare: df[sentence] = df['Case'].apply(lambda x: calculate similarity(x, [sentence])[0]) 将结果保存到新的Excel文件 df.to excel("T12.xlsx", index=False) d. 频次统计: 在输出的xls文档中,用户评论的情感分析与优缺点一致,且SBERT句子相似度大于0.41,那么该优点或缺点的统计频次可以加1。 打开xls表快速排序查看,就能算出各个优点和缺点有几个cases。这个环节相当于人工核验。 情感分析 SBERT句子相似性阈值 GPT归纳优点 POSITIVE 0.41 GPT归纳缺点 NEGATIVE 0.41 报告范例 分析框架 Prompt:根据知识库回答,XREAL AIR AR眼镜有什么优点或缺点。 Response: 根据数据情况,XREAL AIR以屏幕显示质量取胜,可作为多个设备场景的扩展屏幕。不足地方是设备兼容性有待改善,AR功能缺乏。统计频次由SBERT实现。 Advantages 频次 Versatility with Multiple Devices: They work well with various devices like Steam Deck, iPad, MacBook, Samsung Galaxy phones, and more, offering a unique experience for each. 21 High Quality Display: Users appreciate the clear and vibrant display, with no screen door effect and good readability of text. The field of view and pixel density are praised, offering a crisp and immersive experience. 44 Comfortable for Extended Use: The glasses are generally comfortable, with adjustable nose pads and arms, although some users suggested improvements for longer sessions. 13 Enhanced Media Consumption: They excel in providing a private and large screen experience for watching movies, playing games, and multitasking, particularly beneficial in scenarios like flights or public places. 5 Practical Applications: Users have found them useful for productivity (especially with Samsung Dex), gaming, and even mundane tasks like chores or shopping, providing a novel way to interact with digital content while on the move. 10 Good Build Quality: The build quality is considered premium and durable. 幻觉 Disadvantages 频次 Compatibility Issues: Some users faced compatibility problems, especially with non Samsung or older phones, and noted a lack of features when not using Samsung Dex. 68 Limited AR Functionality: The AR experience is basic and lacks features, with some users finding the Nebula app limited and underdeveloped. 62 Physical Discomfort for Some Users: While generally comfortable, the glasses might cause discomfort for those with larger heads or different facial structures, and the weight can be noticeable over long periods. 45 Optical Limitations: Some users reported issues with focus, blurriness on the screen edges, and discomfort with vision. 47 Dependency on External Power Source: Their reliance on an external device for power can be a downside, as it drains the connected device's battery faster. 幻觉 Price Considerations: Some users felt the price was high compared to the technology offered, especially when adding necessary adapters for certain devices. 幻觉 从上面表格可以看出,GPT有一定幻觉问题,如何避免? • 要求根据知识库回答,若信息不足请勿强行回答。 • 做一下mind map,多问几次,核心部位都会覆盖,示例如下: 问题挖掘 Prompt:请详细解释xxx问题,引用用户描述更好。 Response:(无法做SBERT统计频次) Query GPT response Compatibility Issues 1. 对特定设备的依赖性: 尽管Nreal Air与多种设备兼容,但用户体验通常取决于所连接设备的类型和功能。例如,一位用户提到,虽然他的Samsung Note 8并不在支持的手机列表中,但他能够通过安装Nebula应用和利用Samsung DEX的支持来使用眼镜。这表明,某些功能可能仅在特定的设备或操作系统上可用,限制了眼镜的通用性。 2. 特定手机的限制: 另一位用户提到,他尝试使用Samsung Z Flip 4 5G使用眼镜,但由于该手机没有DeX功能,他无法将眼镜用作外部显示器。尽管他能够在配备DeX的Samsung S21 Ultra 5G上成功使用眼镜,但仍然无法使用Nebula应用中的AR空间功能。这反映了即使在兼容设备列表中,也可能存在功能限制。 3. 软件兼容性问题: 有关Nebula应用的反馈显示,不是所有手机都能完全支持该应用。一些用户发现,即使手机在官方的兼容列表中,某些应用功能可能无法正常运行或完全无效。这意味着即使硬件上可以连接,软件兼容性也可能成为限制用户体验的因素。 4. iPhone用户的挑战: 对于iPhone用户来说,使用Nreal Air眼镜需要额外的适配器和步骤。一位用户指出,尽管他发现在iPhone上的体验优于其他设备,但需要使用特定的适配器解决方案来连接眼镜。这不仅增加了成本,也增加了使用过程中的复杂性。 Limited AR Functionality 1. 基本的AR体验: 一位用户强调,Nreal Air眼镜仅提供3自由度(3DOF)的AR体验,而不是6自由度(6DOF)。这意味着眼镜能够追踪头部方向,但无法追踪用户在空间中的位置。虽然Nreal提供了一个更高端的产品,支持6DOF AR,但Nreal Air的AR体验较为基础,主要限于头部方向的追踪,没有手势追踪或环境感知等高级功能。 2. 软件应用的限制: 另一位用户提到,虽然Nebula操作系统的AR体验“很酷”,但它缺乏足够的功能和应用支持,使其不值得使用。即使是支持头部追踪和手机作为指针的功能,也因为应用功能不足而变得不那么实用。用户还指出,即使手机完全兼容,某些应用可能无法正确运行或根本无法使用。 3. 屏幕镜像的限制: 同一位用户还指出,屏幕镜像功能要求手机屏幕始终处于开启状态,这可能会导致更多的电池消耗。而且,将手机放入口袋时可能会意外触碰屏幕,干扰眼镜上的显示。这些限制减少了屏幕镜像功能的实用性。 4. AR空间的交互挑战: 用户反映,在使用Nreal Air的AR空间功能时,遇到了一些交互上的挑战。例如,使用激光指针时经常会丢失,需要长按主按钮才能重置。虽然能同时运行多个屏幕,但用户指出,即使这种多屏幕的交互很惊人,也不清楚它的实用性如何。 Physical Discomfort 1. 长时间佩戴的不适: 一位用户表示,虽然眼镜的鼻垫对于短期佩戴来说还可以接受,但在长时间使用后,他希望鼻垫能更柔软一些,例如使用记忆泡沫或其他材料,而不是硅胶垫。此外,他提到眼镜的耳杆较短,不能很好地围绕耳朵,长时间佩戴后可能会感到不适。 2. 大小和重量问题: 另一位用户评论说,眼镜的大小较大,可能会给一些头型较宽的用户带来压迫感。尽管可以调节屏幕角度来改变压力点,但用户没有发现任何调节屏幕位置(深度或眼间距)的方法。此外,他们还指出,眼镜的重量可能会在鼻梁上留下压痕。 3. 对特定用户群体的不适: 一位用户详细描述了他在视力方面的问题。他指出,虽然他佩戴隐形眼镜,但在使用Nreal Air时,左眼始终无法完全聚焦。他还提到,由于眼镜是“一码合小”,对于眼间距较宽的用户来说,屏幕的外侧(左眼的左侧和右眼的右侧)会出现模糊。这表明眼镜可能并不适合所有人的面部结构。 4. 音频体验的限制: 同一位用户还比较了Nreal Air和其他具有内置扬声器的眼镜,指出尽管Nreal Air具备音频功能,但在低音和最大音量方面略显不足。这意味着长时间佩戴时,用户可能需要使用其他耳机来获得更好的音频体验。 交叉分析 Prompt:根据用户的购买过程,包括购买前,初次使用,后期使用,放弃使用等过程,分别描述用户遇到的问题。 Response: 用户旅程 遇到问题 购买前 • 兼容性担忧: 用户在购买前担心设备的兼容性问题,尤其是非三星或较旧型号手机用户对是否能充分利用眼镜的功能感到不确定。 • 价格考量: 一些用户对眼镜的价格表示担心,尤其是当需要为特定设备购买额外适配器时,感到成本可能过高。 初次使用 • 设置和激活问题: 用户可能在设备的激活和初始设置中遇到问题,尤其是在安装必要的应用程序和调整设置时。 • 初期适应问题: 在初次使用时,用户可能会发现屏幕边缘模糊或对焦点调整有困难,这影响了他们的初体验。 后期使用 • 长时间使用的舒适度: 随着使用时间的增长,用户可能会发现眼镜的重量和尺寸对于长期佩戴来说不够舒适,尤其是头型较大或面部结构不同的用户。 • 电池效率问题: 用户在后期使用过程中可能注意到眼镜对连接设备的电池寿命有显著影响,尤其是在没有内置电池的情况下。 放弃使用 • AR功能的局限性: 一些用户可能会因为AR体验的基础性和Nebula应用的限制而放弃使用眼镜。 • 长期体验不佳: 如果用户长时间使用后仍然感到不适或找不到眼镜的有效用例,可能会选择不再使用这款产品。 用户画像 Prompt:请问满意用户和不满意用户分布有哪些特征? Response: 满意用户的特征 不满意用户的特征 • 技术适应性高: 通常拥有与眼镜兼容的设备,如最新款的三星智能手机或Steam Deck等。 • 对显示质量的高度评价: 对眼镜的高分辨率、清晰度和色彩鲜艳度感到满意,尤其是在媒体消费如观看电影或游戏时。 • 注重实用性: 他们发现眼镜在日常生活中有实际应用,如提高生产力、游戏和做家务等,并且能够在移动中与数字内容互动。 • 重视便携性: 由于眼镜没有内置电池,而是由连接的设备供电,这些用户欣赏眼镜更轻、更便携的特点。 • 适应性强: 能够快速适应AR眼镜的使用,不太受到初始适应问题的影响。 • 技术兼容性问题: 可能使用的设备不在支持列表上,或者不具备如三星Dex等特定功能,导致功能受限。 • 对光学性能的批评: 遇到焦点问题、屏幕边缘模糊或者因设计而导致的视觉不适。 • 长期舒适性问题: 对于长时间佩戴眼镜时的不适,尤其是头部较大或面部结构不符合眼镜设计的用户。 • 电池续航关切: 对眼镜使用过程中消耗连接设备电池的速度感到担忧。 • 对价格的敏感性: 认为眼镜的价格与提供的技术不相匹配,特别是在需要为某些设备购买额外适配器时。 2. 算法管线 def analyze sentiment(text): """ 对文本进行情感分析,长文本将被分割处理。 """ if len(text) <= 512: 假设512是模型的最大长度 return sentiment analyzer(text)[0]['label'] else: chunks = chunk text(text, 512) sentiments = [sentiment analyzer(chunk)[0]['label'] for chunk in chunks] return max(set(sentiments), key=sentiments.count) 返回最常见的情感 def calculate similarity(text, sentences): """ 计算文本与给定句子之间的相似度。 """ case embedding = sbert model.encode(text, convert to tensor=True) sentences embeddings = sbert model.encode(sentences, convert to tensor=True) similarities = util.pytorch cos sim(case embedding, sentences embeddings)[0] return similarities.cpu().numpy() 对每个案例执行情感分析 df['Sentiment'] = df['Case'].apply(analyze sentiment) 对每个案例计算与给定句子的相似度,并将句子作为列标题 for sentence in sentences to compare: df[sentence] = df['Case'].apply(lambda x: calculate similarity(x, [sentence])[0]) 将结果保存到新的Excel文件 df.to excel("T12.xlsx", index=False) d. 频次统计: 在输出的xls文档中,用户评论的情感分析与优缺点一致,且SBERT句子相似度大于0.41,那么该优点或缺点的统计频次可以加1。 打开xls表快速排序查看,就能算出各个优点和缺点有几个cases。这个环节相当于人工核验。 情感分析 SBERT句子相似性阈值 GPT归纳优点 POSITIVE 0.41 GPT归纳缺点 NEGATIVE 0.41 def analyze sentiment(text): """ 对文本进行情感分析,长文本将被分割处理。 """ if len(text) <= 512: 假设512是模型的最大长度 return sentiment analyzer(text)[0]['label'] else: chunks = chunk text(text, 512) sentiments = [sentiment analyzer(chunk)[0]['label'] for chunk in chunks] return max(set(sentiments), key=sentiments.count) 返回最常见的情感 def calculate similarity(text, sentences): """ 计算文本与给定句子之间的相似度。 """ case embedding = sbert model.encode(text, convert to tensor=True) sentences embeddings = sbert model.encode(sentences, convert to tensor=True) similarities = util.pytorch cos sim(case embedding, sentences embeddings)[0] return similarities.cpu().numpy() 对每个案例执行情感分析 df['Sentiment'] = df['Case'].apply(analyze sentiment) 对每个案例计算与给定句子的相似度,并将句子作为列标题 for sentence in sentences to compare: df[sentence] = df['Case'].apply(lambda x: calculate similarity(x, [sentence])[0]) 将结果保存到新的Excel文件 df.to excel("T12.xlsx", index=False) def analyze sentiment(text): """ 对文本进行情感分析,长文本将被分割处理。 """ if len(text) <= 512: 假设512是模型的最大长度 return sentiment analyzer(text)[0]['label'] else: chunks = chunk text(text, 512) sentiments = [sentiment analyzer(chunk)[0]['label'] for chunk in chunks] return max(set(sentiments), key=sentiments.count) 返回最常见的情感 def calculate similarity(text, sentences): """ 计算文本与给定句子之间的相似度。 """ case embedding = sbert model.encode(text, convert to tensor=True) sentences embeddings = sbert model.encode(sentences, convert to tensor=True) similarities = util.pytorch cos sim(case embedding, sentences embeddings)[0] return similarities.cpu().numpy() 对每个案例执行情感分析 df['Sentiment'] = df['Case'].apply(analyze sentiment) 对每个案例计算与给定句子的相似度,并将句子作为列标题 for sentence in sentences to compare: df[sentence] = df['Case'].apply(lambda x: calculate similarity(x, [sentence])[0]) 将结果保存到新的Excel文件 df.to excel("T12.xlsx", index=False) def analyze sentiment(text): """ 对文本进行情感分析,长文本将被分割处理。 """ if len(text) <= 512: 假设512是模型的最大长度 return sentiment analyzer(text)[0]['label'] else: chunks = chunk text(text, 512) sentiments = [sentiment analyzer(chunk)[0]['label'] for chunk in chunks] return max(set(sentiments), key=sentiments.count) 返回最常见的情感 def calculate similarity(text, sentences): """ 计算文本与给定句子之间的相似度。 """ case embedding = sbert model.encode(text, convert to tensor=True) sentences embeddings = sbert model.encode(sentences, convert to tensor=True) similarities = util.pytorch cos sim(case embedding, sentences embeddings)[0] return similarities.cpu().numpy() 对每个案例执行情感分析 df['Sentiment'] = df['Case'].apply(analyze sentiment) 对每个案例计算与给定句子的相似度,并将句子作为列标题 for sentence in sentences to compare: df[sentence] = df['Case'].apply(lambda x: calculate similarity(x, [sentence])[0]) 将结果保存到新的Excel文件 df.to excel("T12.xlsx", index=False) d. 频次统计: 在输出的xls文档中,用户评论的情感分析与优缺点一致,且SBERT句子相似度大于0.41,那么该优点或缺点的统计频次可以加1。 打开xls表快速排序查看,就能算出各个优点和缺点有几个cases。这个环节相当于人工核验。 情感分析 SBERT句子相似性阈值 GPT归纳优点 POSITIVE 0.41 GPT归纳缺点 NEGATIVE 0.41 在输出的xls文档中,用户评论的情感分析与优缺点一致,且SBERT句子相似度大于0.41,那么该优点或缺点的统计频次可以加1。 打开xls表快速排序查看,就能算出各个优点和缺点有几个cases。这个环节相当于人工核验。 情感分析 SBERT句子相似性阈值 GPT归纳优点 POSITIVE 0.41 GPT归纳缺点 NEGATIVE 0.41 情感分析 情感分析 SBERT句子相似性阈值 SBERT句子相似性阈值 GPT归纳优点 GPT归纳优点 POSITIVE POSITIVE 0.41 0.41 GPT归纳缺点 GPT归纳缺点 NEGATIVE NEGATIVE 0.41 0.41 报告范例 分析框架 Prompt:根据知识库回答,XREAL AIR AR眼镜有什么优点或缺点。 Response: 根据数据情况,XREAL AIR以屏幕显示质量取胜,可作为多个设备场景的扩展屏幕。不足地方是设备兼容性有待改善,AR功能缺乏。统计频次由SBERT实现。 Advantages 频次 Versatility with Multiple Devices: They work well with various devices like Steam Deck, iPad, MacBook, Samsung Galaxy phones, and more, offering a unique experience for each. 21 High Quality Display: Users appreciate the clear and vibrant display, with no screen door effect and good readability of text. The field of view and pixel density are praised, offering a crisp and immersive experience. 44 Comfortable for Extended Use: The glasses are generally comfortable, with adjustable nose pads and arms, although some users suggested improvements for longer sessions. 13 Enhanced Media Consumption: They excel in providing a private and large screen experience for watching movies, playing games, and multitasking, particularly beneficial in scenarios like flights or public places. 5 Practical Applications: Users have found them useful for productivity (especially with Samsung Dex), gaming, and even mundane tasks like chores or shopping, providing a novel way to interact with digital content while on the move. 10 Good Build Quality: The build quality is considered premium and durable. 幻觉 Disadvantages 频次 Compatibility Issues: Some users faced compatibility problems, especially with non Samsung or older phones, and noted a lack of features when not using Samsung Dex. 68 Limited

在 小宇宙note 阅读完整内容