Browse Prior Art Database

A Method and System of Generating Comparable Weights to Derive Holistic Network Customer Experience Score

IP.com Disclosure Number: IPCOM000240818D
Publication Date: 2015-Mar-05
Document File: 6 page(s) / 122K

Publishing Venue

The IP.com Prior Art Database

Abstract

The problem addressed in this invention is to determine the weight of each Telco service per customer so that the overall customer experience for each subscriber upon a set of consumed Telco services could be derived based on them and the QoS of each service. The widely adopted approach to this problem in the Telco industry is to use each subscriber’ s purchased service plan to distinguish the importance they place on each service. Drawbacks are two folds. Firstly, the precision is arguable since the purchased service plan doesn’t reflect the real consumption at all cases. Secondly, service plans are at a monthly basis while the requirements of importance of services are flexible, such as, work days, weekends, working time, non working time, etc. The invention addresses it by a data-driven approach, specifically, data normalization to importance modeling. The problem here is not only the value normalization but also the importance modeling. Directly normalizing consumption values (e.g., divided by maximum or threshold) to generate service importance is not applicable since consumptions are dynamic while the service importance is stable. So a large number of original consumption data is used to fit a model to reveal a consumption pattern on each service by subscribers. The latest stable consumption data of each subscriber is used to locate the subscriber’s service index. We use the linear fitting and three cubed curve fitting on the log values on the two axes (X axes indicates the consumption and Y axes indicates the frequency of the consumption). Binning on the consumption data is performed before the log computation. One of the strengths for the model fitting normalization is to handle the outlier easily, this is necessary for handling the normalization of the large volume consumption data rather than the normalization of the normal scale of QoS data. Also, it preserves the range and introduces the dispersion of the original data and provides the knowledge and flexibility to categorization according to business requirements. Thirdly, the consumption data is discovered to closely conform to a power law distribution. An approximation of fitting is used to recover the consumption model. It is composed by two parts of fitting, linear and three cubed curve fitting on the log values of the independent variable and dependent variable.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 46% of the total text.

Page 01 of 6

A Method and System of Generating Comparable Weights to Derive Holistic Network Customer Experience Score

CX,

((CX

CX

                                      )is an important piece of information for Telco operators reflecting the subscriber's perception of the service and recognition on the operator, which causes complains, indicates the churn, and affects the business performance eventually. It draws attentions of Telco operators to have an effective CEM (Customer Experience Management) to keep the competitive edge in the fragmented and highly competitive markets.

QoE is constructed based on QoS (quality of service) which is at the elementary servicing level and can be measured objectively. Some QoS metrics are, for video apps, Time to Start, Playback Gap Ratio, for voice calls, dropped call ratio, etc.

Holistic QoE is an overall view across different services. How to effectively have a holistic QoE for each subscriber according to their perceived elementary servicing quality, is the problem addressed in this disclosure.

The subjective method is doing survey to get an overall view evaluated or synthesized subjectively from human's input (such as, net promoters score). Pros are direct, accurate, while cons are costly, sometimes complex to conduct, and poor scalability. The pseudo-subjective method trains the model based on the redeemed genuine results. Pros are real-time, relatively high accuracy, supported by the theory of statistics and AI. Cons includes that the training procedure is complicated, training data set should be large and accurate. The rule-based method leverage human's knowledge. Pros are fast to launch and the computation is light. Cons are coarse measurement at the individual level, rule's accuracy is always a challenge.

The core idea of this invention is QoS scores at elementary servicing level are weighted to consolidate the holistic customer experience score. Weighting reflects the importance each subscriber places on the services. Weighting is based on heaviness/lightness each subscriber consumes the services. For every service, statistically fitting all subscribers' consumption distribution to the probability model and based on the model to fit a fixed-range consumption index for each subscriber to indicate the relative heavy or light usage on the service. For each subscriber, calculated consumption indexes of all different services are comparable and used to generate weights for the weighted consolidation calculation.

Advantages of the invention includes, it is a purely data-driven approach, and the data is ready; there is no dependency on human input, it is with high scalability; it is at each individual's level, more precise and flexible than the rule-based segmentation level's weighting and scoring.

The core of the invention is to generate comparable measurement results for different service's consumption which are in different units of measurement and also different in consumption ranges. Fig 1. is the overview of the invention idea on how...