System and method for adapting sentiment analysis to user profiles to reduce bias

专利摘要:
Provided is a system and method for adapting sentiment analysis to user profiles to reduce bias in customer or user generated content, specifically a system and method that discounts or adjusts bias in sentiment data based on the channel from which the content was received and/or the demographic of the user. The system includes a means to detect sentiment bias for any product, service, or company across multiple channels of customer data; a means to construct models to quantize bias by specific demographics and channels; and a means to adjust sentiment model output to reduce inflation by biased groups.
公开号:EP3706063A1
申请号:EP20161571.3
申请日:2020-03-06
公开日:2020-09-09
发明作者:Ian Beaver
申请人:Verint Americas Inc；
IPC主号:G06Q30-00

专利说明:
[0001] This application claims priority to U.S. Provisional Patent Application Serial Number 62/814,899, filed March 7, 2019 , which is hereby incorporated by this reference in its entirety as if fully set forth herein. Field
[0002] Embodiments of the present invention relate to a system and method for adapting sentiment analysis to user profiles to reduce bias in customer or user generated content, specifically a system and method that discounts bias based on the channel from which the content was received. Background
[0003] Modern companies use multiple communication channels or platforms to engage with their customers, handle support requests, as well as to gather feedback and monitor brand perception. Within these channels company specific content is generated as customers ask questions and receive answers from employees and representatives over phone calls, e-mails, chat, and social platforms such as Twitter and Facebook.
[0004] An important aspect of customer feedback regardless of channel is their sentiment about the topic. For example, if customers are upset they may use language and tone that will indicate this, even if they do not say outright that the issue is upsetting them. Sentiment analysis aims to identify the polarity (positive or negative) and intensity of certain texts in order to shed light on people's sentiments, perceptions, opinions, and beliefs about a particular product, service, scheme, etc. [1]. By applying sentiment analysis to customer service texts, it is possible to determine if a product or service is upsetting or satisfying customers and to what degree. Conventional or known sentiment analysis procedures use demographic information to categorize the sentiment of reviewers in product or service reviews [3, 4]. BRIEF SUMMARY OF THE DISCLOSURE
[0005] Accordingly, the present invention is directed to the system and method for adapting sentiment analysis to user profiles to reduce bias that obviates one or more of the problems due to limitations and disadvantages of the related art.
[0006] In accordance with the purpose(s) of this invention, as embodied and broadly described herein, this invention, in one aspect, relates to a of reducing sentiment bias in sentiment analysis scores, the method including one or more processing devices performing operations including collecting attributes of users on a per channel basis; gathering demographics and associating the demographics with all observed users; performing sentiment analysis on content on each channel by each users to determine bias in a segment of the content to produce an original sentiment score; determining a sentiment adjustment factor based on the bias; applying the sentiment adjustment factor to the original sentiment score to compensate for the bias; and generating an adjusted sentiment score.
[0007] In another aspect, the invention relates to a system comprising a processing device; and a memory device in which instructions executable by the processing device are stored for causing the processor to collect attributes of users on a per channel basis; gather demographics and associating the demographics with all observed users; perform sentiment analysis on content on each channel by each users to determine bias in a segment of the content to produce an original sentiment score; determine a sentiment adjustment factor based on the bias; apply the sentiment adjustment factor to the original sentiment score to compensate for the bias; and generate an adjusted sentiment score.
[0008] In yet another aspect, the invention relates to a non-transitory computer-readable storage medium having program code that is executable by a processor to cause a computing device to perform operations, the operations comprising: collecting attributes of users on a per channel basis; gathering demographics and associating the demographics with all observed users; performing sentiment analysis on content on each channel by each users to determine bias in a segment of the content to produce an original sentiment score; determining a sentiment adjustment factor based on the bias; applying the sentiment adjustment factor to the original sentiment score to compensate for the bias; and generating an adjusted sentiment score.
[0009] Additional advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The advantages of the invention will be realized and attained by means of the elements and combinations particularly pointed out in the appended claims. It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention, as claimed.
[0010] Further embodiments, features, and advantages of the system and method for adapting sentiment analysis to user profiles to reduce bias, as well as the structure and operation of the various embodiments of the system and method for adapting sentiment analysis to user profiles to reduce bias, are described in detail below with reference to the accompanying drawings.
[0011] It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only, and are not restrictive of the invention as claimed. BRIEF DESCRIPTION OF THE DRAWINGS
[0012] The accompanying figures, which are incorporated herein and form part of the specification, illustrate system and method for adapting sentiment analysis to user profiles to reduce bias. Together with the description, the figures further serve to explain the principles of the system and method for adapting sentiment analysis to user profiles to reduce bias described herein and thereby enable a person skilled in the pertinent art to make and use the system and method for adapting sentiment analysis to user profiles to reduce bias.
[0013] FIG. 1 is flowchart illustrating a method for adapting sentiment analysis to user profiles to reduce bias according to principles described herein. DETAILED DESCRIPTION
[0014] Reference will now be made in detail to embodiments of the system and method for adapting sentiment analysis to user profiles to reduce bias with reference to the accompanying figures The same reference numbers in different drawings may identify the same or similar elements.
[0015] It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit or scope of the invention. Thus, it is intended that the present invention cover the modifications and variations of this invention provided they come within the scope of the appended claims and their equivalents.
[0016] Some customer groups may be more prone to use social media than other groups, and within these groups several factors may influence the predominant sentiment of a product or service on social media leading to bias when determining overall customer sentiments. For example, social media has been shown to be used more by emotionally unstable people regardless of gender, age, race, or life satisfaction [2]. Therefore, traditional customer service channels or platforms, such as chat or e-mail may show a different sentiment than the social media channels for a similar customer demographic. In spite of this, we do not wish to discard social media data altogether as it is an important and prevalent form of customer service. Instead, by considering the demographics of the customer when calculating their sentiment towards a product, service, or company, potential bias can be reduced or "discounted" to provide a more realistic picture of overall customer sentiment across multiple channels.
[0017] To improve the objectivity of information provided by the regression analysis and to reduce computational complexity of the optimization problem, thereby saving computational resources, such as CPU times and memory spaces, rather than merely categorizing sentiment of reviews, the present system and method modifies sentiment models themselves based on the demographics of the customer in any customer service interaction. For example, knowing that 80% of people over the age of 65 dislike product X, the degree of negativity on negative customer interactions for product X can be identified and reduced for any customers over the age of 65. The amount the negativity can be scaled proportionally to the association of customer's demographics to a negative sentiment for the current topic. By scaling or balancing the output of the sentiment model, a more objective customer sentiment level can be obtained, thus improving existing technological processes involving machine-learning techniques.. Note that if a product receives widespread positive or negative attention regardless of demographics, the net information produced by the sentiment model will remain the same (customer feedback is positive or negative, respectively). This method will not flip sentiment polarity; it will reduce the bias to inflate polarity by specific customer groups on specific channels.
[0018] The method provided herein includes the following steps, which may be performed in several parts. The first is to build a demographic profile of the customer base across all channels of customer service, communities, and social media. The second is to group all of the interactions around common products, topics or services. Next, a model is constructed which surfaces any correlations between specific customer attributes and channels and the resulting sentiment polarity. This is repeated for each product, topic, or service. Then, these models are used to scale the sentiments of specific customer demographics when performing sentiment analysis on an individual customer, content, channel, or service as a whole.
[0019] Referring to FIG. 1, first, all known customer attributes are collected on a per channel basis. 100. For channels where the customer identification is known, such as live chat or e-mail, customer demographic information will be also be known, e.g., by the company. For external channels such as social media, profile information customers have entered across social media channels for which they have accounts can be used as a resource for collecting demographic information. Moreover, social media accounts may provide links or contact information for other social media or accessible demographic information. For example, a user on Linkedln may include his/her Twitter and Pinterest account links on his/her Linkedln profile page. By joining or accessing the three linked identified social media accounts, a more complete demographic of that user can be collected/developed. While this example illustrates three social media platforms linked, it can be appreciated that numerous social media or customer profiles can be associated with one another and accessed by the present system to collect demographic data for a user or users. Thus, user/customer demographics are gathered and then associated with all observed users on each channel 110.
[0020] Once user/customer demographics are gathered and associated with all observed users on each channel, traditional sentiment analysis is performed on all content created on each channel by each user. 120. This can be performed in parallel on a distributed compute cluster, as each user and channel may be an independent data point. Once the sentiment is extracted, a regression analysis is performed similar, which will measure the tendency for a particular demographic to have more positive or negative sentiment than the population has a whole. An exemplary regression analysis is provided in Teresa Correa, Amber Willard Hinsley, and Homero Gil De Zuniga. Who interacts on the web : The intersection of users personality and social media use. Computers in Human Behavior, 26(2):247-253, 2010, which is hereby incorporated by reference in pertinent part as if disclosed herein. The regression analysis can be performed at the product or service level and also at the channel level to decide where the bias resides. Using these models, a sentiment profile can be constructed at any resolution desired (i.e., per user, per product, per channel). This sentiment analysis provides an original set of sentiment scores and will illustrate sentiment bias among various user demographics and/or channel. Using the sentiment scores from the regression analysis and the illustrated sentiment bias, an adjustment value (or value by which to "discount" observed bias) based on the user demographic and/or channel (measured bias) can be determined and applied to the feedback obtained from users/customers (e.g., the original sentiment scores). 130/140.
[0021] Once sentiment profiles are constructed, sentiment scores from the original analysis can be adjusted using the measured bias. For example, if young male Twitter users tend to be highly critical of a new product, for instance, but the Twitter population as a whole is not, any tweets originating from young males can be adjusted by a negative bias inflation factor from the model. In doing so, a highly biased population will have less of an inflationary factor when doing sentiment analysis.
[0022] Accordingly, provided herein are a means to detect sentiment bias for any product, service, or company across multiple channels of customer data; a means to construct models to quantize bias by specific demographics and channels; and a means to adjust sentiment model output to reduce inflation by biased groups.
[0023] The present framework may be performed by a computer system or processor capable of executing program code to perform the steps described herein. For example, system may be a computing system that includes a processing system, storage system, software, communication interface and a user interface. The processing system loads and executes software from the storage system. When executed by the computing system, software module directs the processing system to operate as described in herein in further detail, including execution of the cross-entropy ranking system described herein.
[0024] The processing system can comprise a microprocessor and other circuitry that retrieves and executes software from storage system. Processing system can be implemented within a single processing device but can also be distributed across multiple processing devices or sub-systems that cooperate in existing program instructions. Examples of processing system include general purpose central processing units, applications specific processors, and logic devices, as well as any other type of processing device, combinations of processing devices, or variations thereof.
[0025] The storage system can comprise any storage media readable by processing system, and capable of storing software. The storage system can include volatile and non-volatile, removable and non-removable media implemented in any method or technology for storage of information, such as computer readable instructions, data structures, program modules, or other data. Storage system can be implemented as a single storage device but may also be implemented across multiple storage devices or sub-systems. Storage system can further include additional elements, such a controller capable, of communicating with the processing system.
[0026] Examples of storage media include random access memory, read only memory, magnetic discs, optical discs, flash memory, virtual memory, and non-virtual memory, magnetic sets, magnetic tape, magnetic disc storage or other magnetic storage devices, or any other medium which can be used to storage the desired information and that may be accessed by an instruction execution system, as well as any combination or variation thereof, or any other type of storage medium. In some implementations, the store media can be a non-transitory storage media. In some implementations, at least a portion of the storage media may be transitory. It should be understood that in no case is the storage media a propagated signal.
[0027] Throughout this application, various publications may have been referenced. The disclosures of these publications in their entireties are hereby incorporated by reference into this application in order to more fully describe the state of the art to which this invention pertains.[1] Hongzhi Xu, Enrico Santus, Anna Laszlo, and Chu-Ren Huang. Llt-polyu: identifying sentiment intensity in ironic tweets. In Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015), Association for Computational Linguistics, pages 673-678, 2015. [2] Teresa Correa, Amber Willard Hinsley, and Homero Gil De Zuniga. Who interacts on the web : The intersection of users personality and social media use. Computers in Human Behavior, 26(2):247-253, 2010. [3] Dhruv A Bhatt. Sentiment analysis based on demographic analysis, 2014. US Patent App. 13/675,653 . [4] Nicolas Nicolov, William Allen Tuohig, and Richard Hansen Wolniewicz. Automatic sentiment analysis of surveys, 2009. US Patent App. 12/481,398 .
[0028] While various embodiments of the present invention have been described above, it should be understood that they have been presented by way of example only, and not limitation. It will be apparent to persons skilled in the relevant art that various changes in form and detail can be made therein without departing from the spirit and scope of the present invention. Thus, the breadth and scope of the present invention should not be limited by any of the above-described exemplary embodiments, but should be defined only in accordance with the following claims and their equivalents.

权利要求:
Claims (13)
[0001] A method of reducing sentiment bias in sentiment analysis scores, the method including one or more processing devices performing operations comprising:
collecting attributes of users on a per channel basis;
gathering demographics and associating the demographics with all observed users;
performing sentiment analysis on content on each channel by each users to determine bias in a segment of the content to produce an original sentiment score;
determining a sentiment adjustment factor based on the bias;
applying the sentiment adjustment factor to the original sentiment score to compensate for the bias; and
generating an adjusted sentiment score.
[0002] The method of claim 1, the operations further comprising constructing a model of correlations between specific customer attributes, channels and sentiment polarity.
[0003] The method of any of the preceding claims, wherein performing sentiment analysis comprises applying a regression analysis to content created on each channel by each user.
[0004] The method of claim 3, wherein the regression analysis is performed in parallel via a distributed compute cluster.
[0005] The method of claim 4, wherein the parallel performing of the regression analysis is disturbed according to channel.
[0006] The method of any of the preceding claims, further comprising measuring via the sentiment analysis a tendency for a particularly demographic to have a more positive or negative sentiment than a population of users as a whole and wherein the sentiment adjustment factor measures takes into account this tendency.
[0007] A system comprising:
a processing device; and
a memory device in which instructions executable by the processing device are stored for causing the processor to:
collect attributes of users on a per channel basis;
gather demographics and associating the demographics with all observed users;
perform sentiment analysis on content on each channel by each users to determine bias in a segment of the content to produce an original sentiment score;
determine a sentiment adjustment factor based on the bias;
apply the sentiment adjustment factor to the original sentiment score to compensate for the bias; and
generate an adjusted sentiment score.
[0008] The system of claim 7, the memory device further storing therein instructions executable for causing the processor to construct a model of correlations between specific customer attributes, channels and sentiment polarity.
[0009] The system of any of claims 7 and 8, wherein performing sentiment analysis comprises applying a regression analysis to content created on each channel by each user.
[0010] The system of claim 9, wherein the regression analysis is performed in parallel via a distributed compute cluster.
[0011] The system of claim 10, wherein the parallel performing of the regression analysis is disturbed according to channel.
[0012] The system of claim any of claims 7-11, the memory device further storing therein instructions executable for causing the processor to measure, via the sentiment analysis, a tendency for a particularly demographic to have a more positive or negative sentiment than a population of users as a whole and wherein the sentiment adjustment factor measures takes into account this tendency.
[0013] A non-transitory computer-readable storage medium having program code that is executable by a processor to cause a computing device to perform operations of the method of any one of claims 1-6.

类似技术:

---

公开号 | 公开日 | 专利标题

---

Gensler et al.2015|Listen to your customers: Insights into brand image using online consumer-generated product reviews

---

Provost et al.2013|Data Science for Business: What you need to know about data mining and data-analytic thinking

---

Hilmersson2014|Small and medium-sized enterprise internationalisation strategy and performance in times of market turbulence

---

US9635181B1|2017-04-25|Optimized routing of interactions to contact center agents based on machine learning

---

US10891628B1|2021-01-12|Using cognitive computing to improve relationship pricing

---

Walsh et al.2014|Impact of customer‐based corporate reputation on non‐monetary and monetary outcomes: The roles of commitment and service context risk

---

US10909585B2|2021-02-02|Method and system for programmatic analysis of consumer reviews

---

Kumar2008|Managing customers for profit: Strategies to increase profits and build loyalty

---

Breuer et al.2014|Risk aversion vs. individualism: what drives risk taking in household finance?

---

Yoon et al.2016|Keeping the American dream alive: The interactive effect of perceived economic mobility and materialism on impulsive spending

---

US8612293B2|2013-12-17|Generation of advertising targeting information based upon affinity information obtained from an online social network

---

Makrehchi et al.2013|Stock prediction using event-based sentiment analysis

---

US7080027B2|2006-07-18|Method and system for analyzing the effectiveness of marketing strategies

---

JP5450051B2|2014-03-26|Behavioral targeting system

---

Zhu et al.2004|A mathematical model of service failure and recovery strategies

---

US10147037B1|2018-12-04|Method and system for determining a level of popularity of submission content, prior to publicizing the submission content with a question and answer support system

---

US9959548B2|2018-05-01|Method and system for generating social signal vocabularies

---

Moriuchi2019|Okay, Google!: An empirical study on voice assistants on consumer engagement and loyalty

---

US8341102B2|2012-12-25|User state presumption system, user state presumption method, and recording media storing user state presumption program

---

US10296961B2|2019-05-21|Hybrid recommendation system

---

US20130282479A1|2013-10-24|System for generating scores related to interactions with a revenue generator

---

AU2011205137B2|2014-04-17|Social media variable analytical system

---

AU2016346497A1|2018-04-26|Method and system for performing a probabilistic topic analysis of search queries for a customer support system

---

US10366400B2|2019-07-30|Reducing un-subscription rates for electronic marketing communications

---

CN104750674B|2018-12-21|A kind of man-machine conversation's satisfaction degree estimation method and system

同族专利:

---

公开号 | 公开日

---

US20200311348A1|2020-10-01|

引用文献:

---

公开号 | 申请日 | 公开日 | 申请人 | 专利标题

---

法律状态:
2020-08-07| STAA| Information on the status of an ep patent application or granted ep patent|Free format text: STATUS: THE APPLICATION HAS BEEN PUBLISHED |

---

2020-08-07| PUAI| Public reference made under article 153(3) epc to a published international application that has entered the european phase|Free format text: ORIGINAL CODE: 0009012 |

---

2020-09-09| AK| Designated contracting states|Kind code of ref document: A1 Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR |

---

2020-09-09| AX| Request for extension of the european patent|Extension state: BA ME |

---

2021-03-12| STAA| Information on the status of an ep patent application or granted ep patent|Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE |

---

2021-04-14| 17P| Request for examination filed|Effective date: 20210309 |

---

2022-02-10| STAA| Information on the status of an ep patent application or granted ep patent|Free format text: STATUS: EXAMINATION IS IN PROGRESS |

优先权:

---

申请号 | 申请日 | 专利标题

---