北京大学网络大数据管理与应用作业：倒排索引

The use of social robots and IE from the Internet is a hot topic in the field of recommendation and information engineering research. The main advantage of social robots is to help humans in daily life. At present, people are employing social robots in public for different purposes, including education, recommendations, and healthcare activities. For recommendations, the social robot is unable to provide precise information because of the system limitations. Most of these systems are based on predefined words and sentences. However, based on these predefined words and sentences, a social robot cannot extract efficient information from the hazy environment of the Internet. Therefore, extensive methodological work in text mining is needed to automatically get the FTQ from the social robot, re-treat the FTQ, and then retrieve the information based on the query. A variety of research has been done on this issue, and several ideas on text mining for social robots have been proposed [3], [4]. Unified communications and rich web interactions are employed to help a Neel robot connect with humans in an interpersonal manner in a shopping mall. The rich web interaction enables the robot to make links with users and help customers to get deals and offers. However, this system may not recommend a precise item because it trusts the data updated by sellers. Therefore, this system needs preference data. A coupon-giving robot system was developed for advertising in a shopping mall [3]. In this system, two tests show the efficiency of a recommendation: varying conversation schemas, and the existence of a robot. The presence of a robot strongly attracts customers. However, there is a limitation to rules for recommendations, and a category of words is used for speech recognition, which affects the accuracy of results. A text mining–based recommendation system was developed for human–robot interaction [7]. In this system, the robot communicates with the human to get the oral query, and then obtains the information employed in an external corpora. This system recommends a movie according to a user’s oral query. The idea of robot rules–based recommendation for retail shops was proposed [4]. This system tracks the customer’s location and recommends precise products that are near them. The system needs semantic rules to recommend products. Human–robot interaction was examined by using various social activities in a hotel [8]. In this system, one robot detects and greets the customers, and another converses with body gestures, so the guest obtains information by talking. A cloud robotics service is employed in a smart city for emergency management operations [9]. This system uses small unmanned aerial vehicles (UAVs) as agents connected to the network. This network infrastructure allows the UAVs to benefit from storage resources and to control open data from common knowledge. A NAO robot and a smart pen are employed to improve social communications with dementia patients [10]. The NAO robot constantly monitors patient activity and assists the patient in daily life situations. In this system, patients can communicate with the NAO robot by using speech functionality. The NAO robot and multiple cameras are also employed for colored object recognition [11]. This system allows the NAO robot to receive oral commands from users to find a needed colored object. Fuzzy logic is employed to recognize the color based on the user’s perception and multiple cameras to improve the quality of the recognitions. At present, the increase in Internet data makes the recommendation debate more challenging. Most recommendation robot systems are based on predefined words and sentences, which may not allow the robot to extract precise information. Therefore, a full-text query can overcome this issue. The FTQ employs an existing search engine to obtain data about a particular topic or item. However, search engines use a keyword-matching mechanism, and are incapable of extracting the aim of the query from data on their servers [1]. To overcome this problem, an intelligent effort is needed to re-treat the user query and to obtain the required information from the intensively blurred environment of the Internet. A full-text search query is employed to extract information from the Internet [12]. This system showed that an FTQ search performs well when keywords of the query are used in the extracted documents. Its primary concern is precision, not recall. However, there are some limitations in the system; the existing search engines employ keyword-matching mechanisms. Therefore, this approach is inadequate at extracting appropriate data from the heterogeneous sources of Internet information. Currently, sentiment analysis–based recommendation has become a hot topic in research. There are two approaches to sentiment analysis: sentiment classification and feature-level sentiment analysis. The text is categorized (positive, neutral, and negative) in sentiment classification approaches, and the information is intended for manually characterizing the sentiment words [13]. In feature-level sentiment analysis, the features are described to extract sentiment words from text [14]. The idea of summarization and sentiment analysis was presented [15], which defines the feature sentiment (negative or positive) of a product by employing a lexicon-based method. Recently, an ontology has been employed in the area of recommendations, IE, and sentiment analysis. IE is used to transfer natural language text into structured information (Daniel et al., 2008). This transformation is obtained by identifying relevant concepts, relationships, and instances. However, natural language has ambiguity (single words can have many meanings). Therefore, the process of IE is a difficult task. Ontology-based IE overcomes this difficulty by organizing the domain knowledge through an ontology. An ontology is a shared conceptualization of a specific domain through concepts, instances, and relationships, which is in a human-understandable and machine-readable format [1], [16]. A regular ontology is applied to find features in movie reviews [17]. This ontology is suitable for the extraction of a planned set of data. Nevertheless, Internet data are imprecise in structure. Consequently, a regular ontology is inadequate for describing the fuzzy terms of features (e.g., city {clean, average, and dusty}). A regular ontology with fuzzy logic works remarkably well when the input is uncertain. Opinion categorization of online item reviews was suggested to measure the linguistic hedges on sentiment labels [18]. The system automatically extracts opinion phrases from reviews and categorizes the reviews in terms of positive and negative. Moreover, the sentiment scores are stored in linguistic variables and presented in table format. An ontology is employed to store all variables with an opinion score, and to provide a knowledgebase platform for the classification of feature polarity. Opinion mining based on an ontology was suggested to categorize and examine online reviews [19]. The system shows that parts of speech can be different, which leads to ambiguous analysis, and decreases the accuracy of the review classifier. The system employs an SVM as an opinion analyzer to compute the precise measure of words for sentiment analysis. Most of the existing research has its limitations in recommendations, IE, and sentiment analysis. Mostly, the recommendation and IE systems are based on predefined words and a regular ontology, respectively. It is a fact that predefined words–based systems are unable to recommend the correct item, and a regular ontology cannot extract the anticipated result from a blurred resource of data. To the best of our knowledge, the proposed merged ontology and SVM-based recommendation and IE is a first effort to automatically retrieve information and extract the meaning of the data for disabled users. This system can e