Sensing the Enterprise: The Push and Pull of Data Analytics

Sensing the Enterprise: The Push and Pull of Data Analytics

Key words: #TextAnalysis, #ArtificialIntelligence, #Sensors

Traditionally, approaches to data life-cycle management have moved from the bottom, up – what data do we have, and how might we turn it into valuable information???Increasingly, however, the reduced cost of data-storage, coupled with the rise of “big data” as a field of inquiry transecting all industries yields not only more complex analytic methods, but also a new privilege – that of being discriminating.??Particularly with large-scale data, organizations must now consider?if?the data that is being collected is the?right?data at all.??Further, the exponential increases in computer processing capability expand the sensing capacity of “things” further into their environments.??For large bureaucracies to effectively sense and respond to their environment, the colossal stride towards analytic wisdom must comprise both a?push?and a?pull?of data into knowledge, and knowledge into actionable data.??

It is in the digitalization of knowledge that Big Data is born, with all of its breadth and heft. The following research explores three topic areas which exemplify various aspects of the push/pull process of decision-support analytics.??First, a glimpse into text analytics techniques demonstrates how unstructured data may be digitally captured from the knowledge layer to inform an organization’s knowledge of its customer.??Next, the varied materializations of artificial intelligence techniques reach for a simultaneous comprehension of all interacting complexities, which may both signal decision-makers, or take autonomous and/or semi-autonomous self-corrective control of a process.??Finally, with the aid of imbedded sensors, humans extend the digitized perception of their environment through balanced networks of data-processing systems.

Text Analysis

Approximately 80% of today’s enterprise data comprises unstructured data in the forms of documents, e-mails, instant messages, and social-media comments.??Extracting knowledge from unstructured data is a process more complex, ambivalent and time-consuming than that of mining numeric data.??Text analysis remains limited by the computer’s ability to process and understand human language.??Because of this difficulty, the value of unstructured text formats remains largely untapped by organizations.??Nevertheless, some organizations are successfully maturing text analytics, which are predominantly applied for gleaning marketing and customer service insights.??In?Using Text Analytics to Derive Customer Service Management Benefits from Unstructured Data, (Müller et al., 2016) conducted a comparative assessment of the text analytics journeys of three organizations, who variously apply a modern, advanced text analytics tool to the management of customer-service.??

Unlike early natural language processing systems – which were developed by linguists, for linguists – advanced text analytics possess four distinctive features that render text analytics feasible to commercial business applications: (1) the foundation of advanced text analytics tools are?data-driven, rather than rule-based; (2) the speed of text analytics systems is in?real time, rather than in periodic batch runs; (3) the logic of text analytics solutions make?probabilistic?inferences rather than deterministic decisions; and (4) the outputs of text analytic systems are intuitive?visual displays, rather than cryptic annotations.??Despite the differences spanning the organizations’ experiences, products, geographies, responses, etc., each of their journeys followed the same basic structure.??Each began with taking a detailed look at historical text data, continued by monitoring streaming data, and ended with enacting decisions that changed the nature of how work was achieved.??From this shared journey, the research team distilled four general lessons that frame an analytic maturity roadmap for implementing text analytics within an organization: (1) position text analytics in business units, rather than in IT; (2) learn the basics of text analysis by analyzing the past; (3) move eventually toward real-time monitoring of textual data streams in real-time; and (4) embed textual analytics into operational business processes.?

In?Research on the Text Classification Based on Natural Language Processing and Machine Learning, (Keming, 2016) reported on the efficacy of applying natural language processing and machine learning on the performance of text classification.??In an era of rapid network formation, big data, and its repercussive wave of information overload can be mediated through data mining and machine learning techniques.??In particular, text-data mining increasingly emerges as a requirement in the information era.??According to Keming,?automatic text categorization?methods have achieved significant development and application in the field of information management as a key technology for processing and organizing vast quantities of text data.??However, as this method of supervised learning becomes increasingly appealing, text categorization nevertheless requires further maturation in the areas of knowledge acquisition and problems of uncertainty.?

Borrowing from the emerging discipline of Computational Linguistics, Keming assessed the application of Natural Language Processing (NLP) techniques, to (1) the problem of high-dimensional feature space in automatic text categorization – which both increases computational complexity and reduces classification accuracy as the number of categorical variables is increased – and (2) the problem of word segmentation – the accuracy of which impacts a satisfactory classification effect.??Within the field of artificial intelligence, machine learning is one of the most important fields of research by which machines study, acquire and simulate knowledge of mankind’s activities.??Rough sets theory?is an emerging learning method within the field of AI that offers an effective method for reducing uncertain, inconsistent or incomplete attributes and values into accurate knowledge.??Additionally, methods in?term clustering?reduce the typically high-dimensionality of text representational space using various similarity measures to group and map elements of the vector space. The results of a comparative assessment of algorithmic modelling techniques indicates that the application of the above discussed NLP and machine learning techniques to increase both the efficiency and efficacy of text classification produces a competitive edge among modern text categorization technologies.

In?Insights from Hashtag #supplychain and Twitter Analytics: Considering Twitter and Twitter Data for Supply Chain Practice and Research, (Chae, 2015) proposed a novel, analytical framework for analyzing Twitter (TA) data, thereby demonstrating the potential role of social-media data for supply-chain practice and research.??Although businesses and research communities increasingly explore the potential opportunities of applied social media and big data, the field of supply-chain management has remained focused on the application of traditional data sources and analytical techniques.??Despite this, many supply-chain industry leaders recognize the potential value contained in big data, particularly for supply-chain intelligence.??With the broad emergence of Web 2.0 applications, social media platforms largely contribute to the explosive growth of big data sources in recent years.??

As the fastest growing social platform, Twitter lends itself to analytic studies in three ways: (1) quantity – over 270 active users generate over 500 tweets per day; (2) content – users follow product, brands and services, discussing them in Twitter; and (3) access – through the Twitter API, Twitter makes its data “open” to public consumption.??The proposed Twitter analytics framework consists of three types of analytic techniques, which focus on the different dimensions of Twitter data.??First, descriptive analytics (DA) seeks to apply quantitative methods to forming an understanding of the tweets, such as the number of tweets, word counts, user per tweets, number of hashtags, etc.??Second, content analytics (CA) applies natural language processing and data mining techniques (such as term frequency, document clustering and sentiment analysis) to derive information from unstructured data.??Finally, network analytics (NA) performs topological, centrality and community analyses to assesses patterns in user engagements, contained in user replies and re-tweets.??By applying this framework to the analysis of 22,399 #supplychain tweets, Chae provided a proof-of-concept demonstration of the potential use and potential role of social-media data for supply-chain practices.????

Artificial Intelligence

The performance of logistics networks depends upon the system’s ability to capture and accurately report measures of an unstable environment.??Data warehouse technologies support supply-chain decision-making by providing data structures to support analytic tools and by storing data in redundant and aggregate ways to support the speed of interactive analysis.??On-line analytic processing (OLAP) software provides the business user access to the data warehouse, and enables the conversion of data into consumable information.??Enabling a flexible and multidimensional analysis of the data, OLAP’s help to realize key performance measures, which help managers guide logistic systems.??Additionally, the application of “smart” management systems (KPI-MS’s) further enable continual improvements of logistics networks, applying both a predictive alert and prescriptive list of prioritized actions relative to a key performance indicator.??

Together, the data warehouse, OLAP, and KPI-MS form the architectural components of most state-of-the-art logistics decision-support systems (DSS’s).??In?A Reinforcement Learning Approach for a Decision Support System for Logistics Networks (2015), authors Rabe and Dross explored an additional learning component to support?smart?decision-support systems.??Using a discrete-event simulation model to predict the consequences of possible changes in the logistics network, the authors presented a case study for how reinforcement learning (RL) techniques enable the DSS to learn and form knowledge from simulated experiences about possible consequences of actions.??Within the field of Artificial Intelligence, reinforcement learning is a machine learning technique concerned with learning the actions that a software agent?should?take in order to?maximize?rewards.??As the authors suggested, the strength of an RL-integrated DSS is in its capacity to support multiple, autonomous interdependence analyses of actions that improve the overall network situation, rather than each KPI disjointedly.??With the concept of a central Heuristic Unit, the authors’ proposed DSS achieved a holistic support capability supporting development across the entire logistics network.???

Across the three primary economic sectors of industry, transportation and building, conservation of energy is a critical task.??Chou and Bui (2014) ascertained that buildings consume a substantial share of global energy consumption, and that the building sector consumes more than 30% of the total energy, worldwide.??Increasing the efficient energy performance of buildings not only will reduce the demand for energy, it will also help to reduce the high levels of CO2?emissions correlated to the consumption of building energy.??The identified influential parameters substantially affecting building energy consumption may be grouped into two main categories: (1) physical properties; and (2) meteorological conditions.??Together, the dynamic relationship makes the prediction of both heating and cooling loads a challenging task, and thus complicate the engineer’s accurate understanding of a designed building’s forecasted energy performance.

In?Modeling Heating and Cooling Loads by Artificial Intelligence for Energy-efficient Building Design (2014), Chou and Bui performed a comparative assessment of various Artificial Intelligence models in predicting the energy performance of buildings (EPB).??Simulating the human inference process, these inference models extrapolated new facts from historical information, as well as adaptively changed in response to changes in historical data.??With the intent of determining how and whether AI can assist building designers in analyzing the influence of input parameters, the researchers compared the speed and performance of the following five AI techniques in predicting building heating and cooling loads: (1) support vector regression (SVR); (2) artificial neural networks (ANN); (3) classification and regression tree (CART); (4) chi-squared automatic interaction detector (CHAID); and (5) general linear regression (GLR).??Of these data mining techniques, the authors determined that both an ensemble model (SVR + ANN) and SVR are superior in both model robustness and prediction accuracy of the proposed models. The authors concluded with a recommendation that further studies in the application of AI techniques to the problem of predicting building energy performance focus on the use of evolutionary computing methods and swarm intelligence algorithms to further investigate improvements to parameters optimization models predicting heating and cooling load efficiency.

As a new form of business management, the Virtual Enterprise (VE) is a temporarily organized structure that aims at the earliest possible release of new products.??As a networked organization, the VE functions to coordinate the exchange of information between various production units and geographically dispersed objects.??In the VE, transportation arises as the key link in the supply chain supporting efficient material flow, as product assembly depends upon on-time deliveries of all of the components.??Six major factors influence the choice of the required means of transportation: (1) delivery time; (2) cost of transportation; (3) reliable compliance with the cargo delivery schedule; (4) frequency of shipments; (5) ability to transport various cargoes; and (6) ability to deliver the cargo to any destination.??Additionally, a variety of transportation modes, as well as vehicles and logistics partners lends a great deal of uncertainty to the problem of optimizing transportation of material resources.??

To this problem of uncertainty in transport logistics problems, researchers Pavlenko et al. applied artificial intelligence techniques to determine not only the optimal transport type, but also to infer the optimal transport route for the delivery of components supporting VE manufacturing (2017).??Fuzzy sets theory enables analyst both to describe vague concepts and knowledge about the world, as well as to infer new information from this knowledge.??Rather than classifying objects with binary true/false values, fuzzy logic theory applies a?range of truth?to determine if an element of a domain is either in or out of a particular set within the domain.??Thus, fuzzy modelling lends itself to the unique requirements of qualitative assessments, particularly in the presence of unknown or uncertain information. Prioritizing by time to and cost of delivery, Pavlenko et al. generated and tested a fuzzy model from two binary fuzzy relations in order to determine the optimal delivery mode, and to demonstrate a fuzzy inference system determining the optimal delivery route.

Sensors

Through electronic skin, robots are able to sense their surroundings through touch.??Tactile systems thus represent a potentially stimulus to establishing a controlled interaction with humans in a real-world environment.??However, reliable tactile systems still require further development and investigation.??In particular, as the increasingly complex integration of the numerous perceptual components comprising the sensory system requires further development aimed at the interpretation of tactile data supporting the recognition of contact surface properties or of certain qualities of touch perception.??Pattern-recognition methods, which prove to be effective in specific materials classification, must be further extended to report the classification of tough modalities and gestures.??In?Computational Intelligence Techniques for Tactile Sensing Systems (2014), Gastaldo et al. investigated the problem of touch-modality recognition of sensory systems.??

The system must discriminate between the various modalities of physical stimuli interacting with the skin surface, which is further complicated by the bi-dimensional structure of sensation that perceives both pressure as well as time-varying distribution of physical interactions.??In this study, researchers apply specific machine learning techniques to touch-modality recognition, and prove the specific computational advantages for processing sensor data under a tensor-based representation. Deriving a machine-learning system for pattern recognition, the proposed pattern-recognition system is specifically designed to deal with the tensor morphology of the tactile signals.??Thus, this research provides to the study of tactile sensing both a procedure to enhance system generalization ability and an architecture supporting multi-class recognition applications.

In farmland, the growth conditions of crops are mainly affected by soil nutrients; a near-equal distribution of nutrients produces nearly-uniform crop conditions.??As an emerging monitoring technology, intelligent wireless sensor networks (WSN’s) collaboratively and comprehensively obtain target information through a large number of sensors deployed in a target area and in self-organized sensor networks.??The deployment model for sensor nodes is an important aspect determining the quality of WSN service.??The number of nodes that are deployed correlate positively with the coverage of information and the sensory ability of the network.??However, a greater number of nodes also correlates with an increase in both cost and sensory redundancy.??

In?The Node Deployment of Intelligent Sensor Networks Based on the Spatial Difference of Farmland Soil (2015), Liu et al. assessed the application of fuzzy c-means clustering as a method for optimizing the number of deployed sensor nodes collecting soil information.??To ensure an even spatial distribution of soil nutrients at each node, researchers applied a fuzzy c-means clustering algorithm to divide farmland into several areas of nearly-uniform soil fertility using various discriminant functions.??The team validated that the normalized intra-cluster coefficient of variance (NICCV) most accurately provided the correct cluster number for the spatial distributions.??Ultimately, the researchers demonstrated a node deployment method that guaranteed complete information monitoring as well as minimized node-deployment costs.??Their recommendations include further exploration of the various soil type, crop type, and season times as factors affecting crop growth to further improve the network deployment algorithm for a more complex wireless sensor network application.

Likewise, recent developments in unmanned aerial vehicles (UAV’s), artificial intelligence and miniaturized thermal imaging systems represent new opportunities for the inexpensive survey of relatively large areas.??The utility of UAV’s that can perform flight paths autonomously and acquire geo-referenced sensor data is increasingly explored across agricultural and environmental monitoring applications.??Particularly in the application of wildlife monitoring, these developments help to resolve the following challenges that limit the accurate and precise estimates of wildlife populations: (1) large geographic range size; (2) low population densities; (3) inaccessible habitats; (4) elusive behaviors; and (5) sensitivity to disturbance.??In addition to UAV regulations, operational costs and public perception issues, the need to reduce the intensive post-processing efforts required to accurately discriminate image objects is one of the most important restrictions facing a broader application of UAVs.??In?Unmanned Aerial Vehicles (UAVs) and Artificial Intelligence Revolutionizing Wildlife Monitoring and Conservation (2016), Gonzolez et al. developed and applied a video processing pipeline to perform automatic detection and classification to not only discriminate between animal and non-animal objects, but also to obtained a population estimate of a chosen species within a surveyed area.

As a new branch of simultaneous localization and mapping (SLAM) technologies, simultaneous localization and tracking (SLAT) emerges as a problem with potentially wide applications in robotic equipment, sensor networks and smart devices.??Although several proposed adaptive structures aim to exploit the advantages of state filtering and parameter estimation algorithms in solving SLAT problems, several problems contribute to the deteriorated performance of sensor message exchange among SLAT networks.??First, these structures rely on specific communication network topologies that fail in general applicability to arbitrarily connected networks.??Second, there is a high computational cost associated with belief propagation in a Bayesian filtering framework.??Third, the structural reliance on distance measurements are rendered less applicable by the functionally limited sensing range of individual sensors.??

In?A Sensor Self-aware Distributed Consensus Filter for Simultaneous Localization and Tracking (2016), Jiang et al. considered the structural design of the brain’s decision mechanism as a framework supporting SLAT problems.??Modelled on the brain, which must often make the best decisions possible with limited and continuously updating information, the research team proposed a new self-aware census filter, which uses a cognitively distributed algorithm to amass and judge evidentiary data, and update decisions as the information is refined.??Like neurological “cells”, sensors within a network carry information including position, velocities etc., while the information exchange circuits collect this information as “evidence”.??If the information exchanged among the network passes a critical threshold, a maximum likelihood estimator makes a decision of the target state, which can be further updated and refined by a feed of new information.??Inimitably, this proposed framework (1) presents a fully distributed adaptive filter for all sensors information delivery within a network, (2) features a likelihood parameter estimation algorithm for making and updating decisions, and (3) reduces computation cost and complexity by removing unnecessary communication overheads.

Discussion

????????????In an adaptive enterprise, how organizations make sense of data is merely one half of the battle; businesses must also integrate decisions about?what?data is needed to inform?which?decisions.??The research described herein explore a sample of the emerging “pull” techniques from across a variety of industries, which serve to illustrate that the value of data analysis is less about knowing the past, and more about shaping better responses to the present.???Text-mining as a technique can provide an essential “listening” capability, enabling an organization to better understand the needs of its potential and existing customers, partners and employees.??Artificial Intelligence – specifically, machine learning techniques and “learning” designs – help to enable human decisions through autonomic simulations of multiple actions, and through holistic optimization of multiple factors, simultaneously.??In the form of sensors, “intelligent” robotics expands the machine’s ability both to sense and codify knowledge from the external environment into data, and to process that data in near real-time.??Together, text-mining, machine learning, and sensor processing all aim to mature analytics into intelligence better enabling human decisions.?

Works Cited

A Reinforcement Learning approach for a Decision Support System for logistics networks. (2015).?2015 Winter Simulation Conference (WSC), Winter Simulation Conference (WSC), 2015, 2020. doi:10.1109/WSC.2015.7408317

Chae, B. (2015). Insights from hashtag #supplychain and Twitter Analytics: Considering Twitter and Twitter data for supply chain practice and research.?International Journal of Production Economics,?165247-259. doi:10.1016/j.ijpe.2014.12.037

Chou, J., & Bui, D. (2014, October 1). Modeling heating and cooling loads by artificial intelligence for energy-efficient building design.?Energy & Buildings,?82, 437-446. doi:10.1016/j.enbuild.2014.07.036

Gastaldo, P., Pinna, L., Seminara, L., Valle, M., & Zunino, R. (2014, June 19). Computational intelligence techniques for tactile sensing systems.?Sensors (Basel, Switzerland),?14(6), 10952-10976. doi:10.3390/s140610952

Gonzalez, L. F., Montes, G. A., Puig, E., Johnson, S., Mengersen, K., & Gaston, K. J. (2016, January 14). Unmanned Aerial Vehicles (UAVs) and Artificial Intelligence Revolutionizing Wildlife Monitoring and Conservation.?Sensors (Basel, Switzerland),?16(1). doi:10.3390/s16010097

Jiang, X., Ren, P., & Luo, C. (2016, October). A sensor self-aware distributed consensus filter for simultaneous localization and tracking.?Cognitive Computation,?8(5), 828-838. doi:10.1007/s12559-016-9423-7

Keming, C. (2016, July). Research on the text classification based on natural language processing and machine learning.?Journal of the Balkan Tribological Association,22(2A-I), 1915-1924. Retrieved July 7, 2017, from Academic Search Complete.

Müller, O., Debortoli, S., Junglas, I., & vom Brocke, J. (2016). Using Text Analytics to Derive Customer Service Management Benefits from Unstructured Data.?MIS Quarterly Executive,?15(4), 243-258.

Liu, N., Cao, W., Zhu, Y., Zhang, J., Pang, F., & Ni, J. (2015, November 11). The Node Deployment of Intelligent Sensor Networks Based on the Spatial Difference of Farmland Soil.?Sensors (Basel, Switzerland),?15(11), 28314-28339. doi:10.3390/s151128314

Pavlenko, V., Pavlenko, T., Morozova, O., Kuznetsova, A., & Voropai, O. (2017, April). Solving transport logistics problems in a virtual enterprise through artificial intelligence methods.?Transport Problems: An International Scientific Journal,12(2), 31-42. doi:10.20858/tp.2017.12.2.4